chore: bump version to 0.13.4

* docs(concepts): document the spec-of-specs feature breakdown approach (#3423)

Add a dedicated "Spec of Specs" concept page describing how to decompose a
large feature into a roadmap of smaller, independently-specified sub-features
using the existing Spec Kit flow. Covers the roadmap pass, the roadmap
artifact template, specifying each sub-feature, bidirectional sub-spec/roadmap
linking, keeping them in sync, a worked example, and optional automation.

Link the new page from the "Handling Complex Features" decomposition section
and add it to the docs table of contents.

Closes #3423

Assisted-by: GitHub Copilot (model: Claude Opus 4.8, autonomous)
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Copilot-Session: f236bc38-a7c6-4063-a79c-6ba81aa685b5

* Potential fix for pull request finding

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

---------

Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

.devcontainer/post-create.sh

@@ -97,6 +97,17 @@ echo -e "\n🤖 Installing CodeBuddy CLI..."
 run_command "npm install -g @tencent-ai/codebuddy-code@latest"
 echo "✅ Done"
 
+echo -e "\n🤖 Installing Factory Droid CLI..."
+run_command "npm install -g droid@latest"
+
+if ! command -v droid >/dev/null 2>&1; then
+  echo -e "\033[0;31m[ERROR] Droid CLI installation did not create 'droid' in PATH.\033[0m" >&2
+  exit 1
+fi
+
+run_command "droid --version > /dev/null"
+echo "✅ Done"
+
 # Installing UV (Python package manager)
 echo -e "\n🐍 Installing UV - Python Package Manager..."
 run_command "pipx install uv"

.github/ISSUE_TEMPLATE/agent_request.yml

@@ -8,7 +8,7 @@ body:
       value: |
         Thanks for requesting a new agent! Before submitting, please check if the agent is already supported.
 
-        **Currently supported agents**: Amp, Antigravity, Auggie CLI, Claude Code, Cline, CodeBuddy, Codex CLI, Cursor, Devin for Terminal, Firebender, Forge, Gemini CLI, GitHub Copilot, Goose, Grok Build, Hermes Agent, IBM Bob, Junie, Kilo Code, Kimi Code, Kiro CLI, Lingma, Mistral Vibe, Oh My Pi, opencode, Pi Coding Agent, Qoder CLI, Qwen Code, RovoDev ACLI, SHAI, Tabnine CLI, Trae, ZCode, Zed
+        **Currently supported agents**: Amp, Antigravity, Auggie CLI, Claude Code, Cline, CodeBuddy, Codex CLI, Cursor, Devin for Terminal, Factory Droid, Firebender, Forge, Gemini CLI, GitHub Copilot, Goose, Grok Build, Hermes Agent, IBM Bob, Junie, Kilo Code, Kimi Code, Kiro CLI, Lingma, Mistral Vibe, Oh My Pi, opencode, Pi Coding Agent, Qoder CLI, Qwen Code, RovoDev ACLI, SHAI, Tabnine CLI, Trae, ZCode, Zed
 
   - type: input
     id: agent-name

.github/ISSUE_TEMPLATE/bug_report.yml

@@ -71,6 +71,7 @@ body:
         - Codex CLI
         - Cursor
         - Devin for Terminal
+        - Factory Droid
         - Firebender
         - Forge
         - Gemini CLI

.github/ISSUE_TEMPLATE/feature_request.yml

@@ -65,6 +65,7 @@ body:
         - Codex CLI
         - Cursor
         - Devin for Terminal
+        - Factory Droid
         - Firebender
         - Forge
         - Gemini CLI

.github/aw/actions-lock.json

@@ -1,10 +1,30 @@
 {
   "entries": {
+    "actions/checkout@v6.0.3": {
+      "repo": "actions/checkout",
+      "version": "v6.0.3",
+      "sha": "df4cb1c069e1874edd31b4311f1884172cec0e10"
+    },
+    "actions/download-artifact@v8.0.1": {
+      "repo": "actions/download-artifact",
+      "version": "v8.0.1",
+      "sha": "3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c"
+    },
     "actions/github-script@v9.0.0": {
       "repo": "actions/github-script",
       "version": "v9.0.0",
       "sha": "3a2844b7e9c422d3c10d287c895573f7108da1b3"
     },
+    "actions/setup-node@v6.4.0": {
+      "repo": "actions/setup-node",
+      "version": "v6.4.0",
+      "sha": "48b55a011bda9f5d6aeb4c2d9c7362e8dae4041e"
+    },
+    "actions/upload-artifact@v7.0.1": {
+      "repo": "actions/upload-artifact",
+      "version": "v7.0.1",
+      "sha": "043fb46d1a93c77aae656e7c1c64a875d1fc6a0a"
+    },
     "github/gh-aw-actions/setup@v0.79.8": {
       "repo": "github/gh-aw-actions/setup",
       "version": "v0.79.8",

.github/scripts/check_security_requirements.py

@@ -0,0 +1,115 @@
+"""Check that committed security audit requirements are up to date."""
+
+from __future__ import annotations
+
+import os
+import subprocess
+import sys
+from pathlib import Path
+
+
+REPO_ROOT = Path(__file__).resolve().parents[2]
+COMMITTED_REQUIREMENTS = REPO_ROOT / ".github" / "security-audit-requirements.txt"
+DEPENDENCY_INPUTS = ("pyproject.toml", ".github/security-audit-requirements.txt")
+
+
+def _dependency_diff_refs() -> tuple[str, str]:
+    base_ref = os.environ.get("DEPENDENCY_DIFF_BASE", "").strip()
+    head_ref = os.environ.get("DEPENDENCY_DIFF_HEAD", "").strip() or "HEAD"
+    if base_ref and not set(base_ref) <= {"0"}:
+        return base_ref, head_ref
+    # Fallback when no usable base is supplied (push with an all-zero
+    # ``github.event.before``, manual dispatch, etc.). ``HEAD^`` fails on a
+    # shallow checkout or a single-commit repo; that ``git diff`` error is
+    # caught by the caller and deliberately treated as "inputs changed" so the
+    # audit runs anyway — failing safe (audit) rather than skipping silently.
+    return "HEAD^", "HEAD"
+
+
+def _dependency_inputs_changed() -> bool:
+    base_ref, head_ref = _dependency_diff_refs()
+    try:
+        result = subprocess.run(
+            [
+                "git",
+                "diff",
+                "--name-only",
+                base_ref,
+                head_ref,
+                "--",
+                *DEPENDENCY_INPUTS,
+            ],
+            check=True,
+            cwd=REPO_ROOT,
+            stderr=subprocess.PIPE,
+            stdout=subprocess.PIPE,
+            text=True,
+        )
+    except subprocess.CalledProcessError as exc:
+        print(
+            "Could not determine changed dependency inputs; checking requirements.",
+            file=sys.stderr,
+        )
+        if exc.stderr:
+            print(exc.stderr.strip(), file=sys.stderr)
+        return True
+
+    changed_inputs = [line for line in result.stdout.splitlines() if line]
+    if not changed_inputs:
+        print("Dependency audit inputs unchanged; sync check skipped.")
+        return False
+
+    print(f"Dependency audit inputs changed: {', '.join(changed_inputs)}")
+    return True
+
+
+def main() -> int:
+    if not _dependency_inputs_changed():
+        return 0
+
+    generated_requirements_env = os.environ.get("GENERATED_REQUIREMENTS", "").strip()
+    if not generated_requirements_env:
+        print(
+            "GENERATED_REQUIREMENTS must be set to the temporary output file path.",
+            file=sys.stderr,
+        )
+        return 1
+
+    generated_requirements = Path(generated_requirements_env)
+    generated_requirements.parent.mkdir(parents=True, exist_ok=True)
+
+    subprocess.run(
+        [
+            "uv",
+            "pip",
+            "compile",
+            "pyproject.toml",
+            "--extra",
+            "test",
+            "--universal",
+            "--upgrade",
+            "--generate-hashes",
+            "--quiet",
+            "--no-header",
+            "--output-file",
+            str(generated_requirements),
+        ],
+        check=True,
+        cwd=REPO_ROOT,
+    )
+
+    committed = COMMITTED_REQUIREMENTS.read_text(encoding="utf-8")
+    generated = generated_requirements.read_text(encoding="utf-8")
+    if committed == generated:
+        return 0
+
+    print(
+        "Regenerate .github/security-audit-requirements.txt with the documented "
+        "uv pip compile command.",
+        file=sys.stderr,
+    )
+    return 1
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

.github/security-audit-requirements.txt

@@ -0,0 +1,253 @@
+annotated-doc==0.0.4 \
+    --hash=sha256:571ac1dc6991c450b25a9c2d84a3705e2ae7a53467b5d111c24fa8baabbed320 \
+    --hash=sha256:fbcda96e87e9c92ad167c2e53839e57503ecfda18804ea28102353485033faa4
+    # via typer
+click==8.4.2 \
+    --hash=sha256:9a6cea6e60b17ebe0a44c5cc636d94f09bd66142c1cd7d8b4cd731c4917a15f6 \
+    --hash=sha256:e6f9f66136c816745b9d65817da91d61d957fb16e02e4dcd0552553c5a197b76
+    # via specify-cli (pyproject.toml)
+colorama==0.4.6 ; sys_platform == 'win32' \
+    --hash=sha256:08695f5cb7ed6e0531a20572697297273c47b8cae5a63ffc6d6ed5c201be6e44 \
+    --hash=sha256:4f1d9991f5acc0ca119f9d443620b77f9d6b33703e51011c16baf57afb285fc6
+    # via
+    #   click
+    #   pytest
+    #   typer
+coverage==7.15.2 \
+    --hash=sha256:075560438765b7a2ef43bf7aa7758661b53d889df47f062a31bda6c1ade553a2 \
+    --hash=sha256:0901cfe6c13bcd2302da4f83e884555d2a22bda6e4c476f09ef204ba20ca536e \
+    --hash=sha256:094dd37f3ef7b2da8b068b583d1f4c40f91c65197e16c52a71962d5d537fc5db \
+    --hash=sha256:09f5c6ec5901f667bd97dd140b5b9a2586b10efec66f46fb1e6d8135f8b95bdf \
+    --hash=sha256:0e55510bc98ae943cece9e667a6c0fe94c6a92913720dea34243657a17993d0c \
+    --hash=sha256:1121caa19159a38b5463eaae4b1e1fde81e525b15ecc5e000cd5b1a108f743a8 \
+    --hash=sha256:1268ac8fb9ddcd783d3948dbabaf80a5d53bfdaa0575e873e2139a692f797443 \
+    --hash=sha256:1473b3ba8e7ee0f076117b1a72c23f579a2b9e2bb742f48a8d86ea27ca93f91a \
+    --hash=sha256:17c432b5f73ad52ef46fb06019f6fa7c66ce381961cf0f7dfd1d3a4bd3a98145 \
+    --hash=sha256:1adac78e5abc7c5438f7a209c9ca69d06542f0bf481d728b6989ea80b813fdf9 \
+    --hash=sha256:1cd7a5beb7af3e864a13b1f0fb26efd3695da43ef0daf71e586adfffaf34d5b2 \
+    --hash=sha256:1d16e3a7104ea84f03e614611b3edbf6fb6892554b3ab0fe7fbb3f2b2ef04376 \
+    --hash=sha256:25fd15dd40a0a2c51a500d664ca29053c09c3259d998407bf982b6e114696138 \
+    --hash=sha256:2617f8799d268fabdeef42a7e89ac3a23e1deee9025427db2df970f99a89a578 \
+    --hash=sha256:26c3b04a6377fd7c09800921fa934e3a17c0020439cd59df73e73ae1d4b6a78c \
+    --hash=sha256:29c052f7c83ccfcc5c577eaae025d2e4a9bb80daf03c0ac31c996e83b000ce88 \
+    --hash=sha256:2f1ec6f304b156669cfde653b4e9a953f5de87e247ea02ac599bce0ab2744036 \
+    --hash=sha256:2fbeeeecea279727f8ac16c8e1133ddfeee793e985c86ae343d6a5ce744eef8c \
+    --hash=sha256:2ff08701be2d1556fc78b326c80a3e8042da09352ecb3819105f8e386c8a3071 \
+    --hash=sha256:38c9518b7103826c403a461544e3c2e77151e8676d06eaed85911a97e962584a \
+    --hash=sha256:3df60dc267f0a2ca23cb7a9ab1109c62b9335ffbf519fcfe167157c28c09b81d \
+    --hash=sha256:3ed010aa1b69cda8e827aabfca9866216c980e2dca82ab9a78c5f83689964c8b \
+    --hash=sha256:40f633c5c5fc783732f6312280122e859538fa24461235597c13d803ea9a108a \
+    --hash=sha256:42ec3d989421b174a2ab607c1539f24127ad362757b7f1c0c0d7a2993f7eb37b \
+    --hash=sha256:434e68d531858205895eb0d74b73d20b84260de426387d53c422a5acda2cf050 \
+    --hash=sha256:44826758cfe73fcd0e6af5deb4ba6d5417cc1d13df3acb35c93484a11160f846 \
+    --hash=sha256:4510fb9cdf6bb02dfa6af0be4a534b8102d086e22e4a33f8836df663da3d660d \
+    --hash=sha256:48ccc6395958eda89093ecdc35644c86f23a8b23a7f4d44958812b721aad67c1 \
+    --hash=sha256:4d3361879d736f469f45723c11ea1a5bbdaf1f6928f0e632c940378b5aa9b660 \
+    --hash=sha256:582edc45c2040543fef83341be23c43024a3ab3ae0c2d8bc498a06282905ad40 \
+    --hash=sha256:63022c4c8dec1d0342f05c3ede99842fe3d007689acc45e86f123a1746e4a026 \
+    --hash=sha256:67d7602480a47bdf5b675635403625553ebaa70d5a62a657c035149fd401cea0 \
+    --hash=sha256:68af907f595ab01a78f794932ff3bdf929c316d3000810d38dbc247129e26f8b \
+    --hash=sha256:6aa28cfb6488e5453b5b762d65f73aa586380f6693a04d58078ce228a29b06c0 \
+    --hash=sha256:6c0be82b4d4aa5b2704e08518e2252f3e3d110164bcca826816801052e48a7aa \
+    --hash=sha256:6f6966fc30e6f06ca8f98fb0ce51eda6b111b3ee8d066a8b1ec9e77fa06ab55d \
+    --hash=sha256:6fc448c377d6eeb00a47c673494bd9bae29280ca53987e1869e67ebedfe20658 \
+    --hash=sha256:728a33676d4c3f0db977990a4bd421dcaa3be3e53b5b6273036fff6666008e89 \
+    --hash=sha256:7466cc7ab6dc0db871d264bf99e8779f0917ee63d40730af0552f71535a6e072 \
+    --hash=sha256:77f091ea3a9cc611cd29f433565476bc1936c084ac8eee00ea0e7e70c27e4199 \
+    --hash=sha256:77f0ef5011df53a4bd1b35211ab122287f8d9b8d7aa1c4553e5c2deb24b1d446 \
+    --hash=sha256:7c63387e21ab21f512c69c9756a8c7dadd322c7275edb064064433c9a09c3743 \
+    --hash=sha256:7d29ca7bd67af6e12e74632d65f026eabc1364da5c254494cd914446a28a3ef7 \
+    --hash=sha256:7dc2950a2992cd676d35c20ae63522836deeb034f08874699d14068710af3dc1 \
+    --hash=sha256:7e8f27131dc7cd53de2c137dd207b3720919320b3c20d499dc30aa9ee6173287 \
+    --hash=sha256:81f382c5a94b434ec1f6da607edb904c76d7212e618cd4d1bc9f97bed4120ef5 \
+    --hash=sha256:835ec4e20b45f0a7f63ed78f94065aca00de033403df8377bfe8b9c6abc0a7be \
+    --hash=sha256:8bb9f4b4279187560796a4cdaca3b0a93dd97e48ee667df005f4ed9a97403688 \
+    --hash=sha256:8c726b232659cbd2ae57ade46509eb068c9bd7a06df9fcbff6fe484870006934 \
+    --hash=sha256:913b6c56e110da40e035bbd168353bf7aaa2544a5eaccea5d98a4629aac156c7 \
+    --hash=sha256:97a5c5457a9fb1d6c4e06cfb5dc835871fbfb6a6a51addc9e925bdeff5ef7440 \
+    --hash=sha256:9854ca62c152874b2060772503535be2e8f53f70b8aaa7686b094888d872f984 \
+    --hash=sha256:9911f31aad8906abe337c271343485cf20df5e70df5d2f57f9f136e7b55f26bc \
+    --hash=sha256:9b5bd92ff1ec22e535eab0de75fa6db021992791f461a2aceb7822c625a1187d \
+    --hash=sha256:9deddf09eecb717b7f980414b43d90a5b22ff3967d2949ab29cb0aa83d9e9098 \
+    --hash=sha256:9e36686f7a442185db2400b3df171aac520869faf9deb59df687d28659eda2a6 \
+    --hash=sha256:9f4432898c4bf2fba0435bbe35dd4437d7264565e5a88a21f5b49d8662a6b629 \
+    --hash=sha256:a0f47002c6eeb7c280228467a4cb0cc15ca2103a8421b986b2d3ec04a0f9bd8b \
+    --hash=sha256:a164b50081fc7357331c4024ef4d17b78ba325f8380d05f5a69599a7e05257ee \
+    --hash=sha256:a29ec5305a7335aacee2d799e3422e91e1c8a12474986e2b3b07e315c91be82f \
+    --hash=sha256:a300c6934e0989c327b9e8a1e110329da4641149f872bbe9f70168be66da76c1 \
+    --hash=sha256:a4c46b247b5d4b78f613bd89fea926d32b25c6cc61a50bd1e99ba310348f3dad \
+    --hash=sha256:a638db90c61cd219aeee65e83a24fdaa57269a741ae0cf773309208ac862cee3 \
+    --hash=sha256:a63b9e190711134d581c4d703df5df09851b1acf99792c7aacbbe9f41f0283c9 \
+    --hash=sha256:aaccad4129d735a8a4d526f26929894c9a4e8ef7034566f210b176749d6906e3 \
+    --hash=sha256:ae901f7e55ba405c84ee1cab3d3e962e4e871e4a2bcb9c90911adbd69b42ac5a \
+    --hash=sha256:afa29e2eff3d5729267e2cb2fd4ce9d61c952932fb2694e34ccb5d9540c6a296 \
+    --hash=sha256:affd532502d34c0472d0cdb181325c89f1d2c44992fef0c17e88e7b1576259a1 \
+    --hash=sha256:b171bdd71cb7ff792bf32e376173b0ace7e7963e7e57c58dfc42063a6a7174cd \
+    --hash=sha256:b868acc62aa5de3be7a9d05c2333bf8359ca987e43f9cb30ff8fbda6a024ab73 \
+    --hash=sha256:b9a6367e4aff723e8ee8190836836124284e8fcd4265e307c844010cfa074f3f \
+    --hash=sha256:bbc808daf4f5cd567af8075ecc72d21c6dfef9a254709a621a84c217c935ebc0 \
+    --hash=sha256:bbf44513ceb1589e31948e20eafbde9deaface90e1a1afa5f5f77b4423d17ce6 \
+    --hash=sha256:bcc0aae933921d03096f53b0b03eeb702129fd406dee59f08d2efacc68681fa5 \
+    --hash=sha256:bfd341ccf78128e72c094bc70cc25b3ef309c33c7c2c66ba3ed4309549e02de1 \
+    --hash=sha256:c6a98d698f9e2c8008d0370ec7fc452ebfcc530002ae2d0061170d768b992589 \
+    --hash=sha256:cb0fddaa6884be6aae36ced9544b5e90f7d5f03845a2853bf47a14953a4e8688 \
+    --hash=sha256:cee0f89f4767a6057c8fbf168f8135f18be651300496086bd873e3189fed0487 \
+    --hash=sha256:d17d7512151fedfcc64c1821a8977fc9be0dbf495754669afcab7b57abc98ae9 \
+    --hash=sha256:d46e62cb35d91e6e2589fda6d28074426b0e276422b5d2ebef2c6b11dc60dbfd \
+    --hash=sha256:d50dd325e18ec25bfcc10cd7f99b04df1ab9ec76b0918c260e60817ad0643dee \
+    --hash=sha256:db9c8438057e5b0f6a22a0af99c0c1d26b57fbbdbd1be5861ddb8f897fcc3a2d \
+    --hash=sha256:dee88b1ed88587abd8c0269a1fc1f4cc77f7750d1dfde2869e2a123af420e67d \
+    --hash=sha256:dfd3db045e95960ae3683059571e597fda7cc610106a8916f77c5839048c1deb \
+    --hash=sha256:e26ff680768b8095e8874aabe0e9d3a47a2a9f176a8340d05f8604c56457c23a \
+    --hash=sha256:e370c12133095ff18432de8c044962be85a5a96d90c6fcbce8e17e76236d2328 \
+    --hash=sha256:e38def96ad59853824c97953fdcd2c320a84ba3ce99b417db78af8bb6c3db635 \
+    --hash=sha256:e8f91bce78e32343af184c3b7fa28fcf5a9e2641f4b6623d392038f804939188 \
+    --hash=sha256:eb6bcae8d1a9d305351ecb108232441d11c5cfe9de840a04388ba5d2db8d735c \
+    --hash=sha256:f653e5d7248c1191ec988a85c72edeab46c3ff44f90639a4ed4874ec0be90243 \
+    --hash=sha256:fe41909c9515c3bfdb5f02c4d1f857dba322d9a9a1178069b91eea77889df63a
+    # via pytest-cov
+iniconfig==2.3.0 \
+    --hash=sha256:c76315c77db068650d49c5b56314774a7804df16fee4402c1f19d6d15d8c4730 \
+    --hash=sha256:f631c04d2c48c52b84d0d0549c99ff3859c98df65b3101406327ecc7d53fbf12
+    # via pytest
+json5==0.15.0 \
+    --hash=sha256:56636a30c0e8a4665fe2179c0212f32eae3796dea89ea6f649b9436ecdb39618 \
+    --hash=sha256:7424d1f1eb1d56da6e3d70643f53619862b4ce81440bdb8ecfd6f875e5ba4a71
+    # via specify-cli (pyproject.toml)
+markdown-it-py==4.2.0 \
+    --hash=sha256:04a21681d6fbb623de53f6f364d352309d4094dd4194040a10fd51833e418d49 \
+    --hash=sha256:9f7ebbcd14fe59494226453aed97c1070d83f8d24b6fc3a3bcf9a38092641c4a
+    # via rich
+mdurl==0.1.2 \
+    --hash=sha256:84008a41e51615a49fc9966191ff91509e3c40b939176e643fd50a5c2196b8f8 \
+    --hash=sha256:bb413d29f5eea38f31dd4754dd7377d4465116fb207585f97bf925588687c1ba
+    # via markdown-it-py
+packaging==26.2 \
+    --hash=sha256:5fc45236b9446107ff2415ce77c807cee2862cb6fac22b8a73826d0693b0980e \
+    --hash=sha256:ff452ff5a3e828ce110190feff1178bb1f2ea2281fa2075aadb987c2fb221661
+    # via
+    #   specify-cli (pyproject.toml)
+    #   pytest
+pathspec==1.1.1 \
+    --hash=sha256:17db5ecd524104a120e173814c90367a96a98d07c45b2e10c2f3919fff91bf5a \
+    --hash=sha256:a00ce642f577bf7f473932318056212bc4f8bfdf53128c78bbd5af0b9b20b189
+    # via specify-cli (pyproject.toml)
+platformdirs==4.11.0 \
+    --hash=sha256:0555d18370482847566ffabcaa53ad7c6c1c29f195989ae1ed634a05f76ea1e0 \
+    --hash=sha256:360ccded2b7fce0af0ff80cc8f5942a1c5d99b0e856033acb030bfc634709e74
+    # via specify-cli (pyproject.toml)
+pluggy==1.6.0 \
+    --hash=sha256:7dcc130b76258d33b90f61b658791dede3486c3e6bfb003ee5c9bfb396dd22f3 \
+    --hash=sha256:e920276dd6813095e9377c0bc5566d94c932c33b27a3e3945d8389c374dd4746
+    # via
+    #   pytest
+    #   pytest-cov
+pygments==2.20.0 \
+    --hash=sha256:6757cd03768053ff99f3039c1a36d6c0aa0b263438fcab17520b30a303a82b5f \
+    --hash=sha256:81a9e26dd42fd28a23a2d169d86d7ac03b46e2f8b59ed4698fb4785f946d0176
+    # via
+    #   pytest
+    #   rich
+pytest==9.1.1 \
+    --hash=sha256:1088fbde8f2b49d95a549a195707afa7a76a3ce9bcadc26b6d71f0ffda5fe313 \
+    --hash=sha256:37a86b45efb9a47a61a36449063e8e18d0cab3161329fc099eb21783169c4f0c
+    # via
+    #   specify-cli (pyproject.toml)
+    #   pytest-cov
+pytest-cov==7.1.0 \
+    --hash=sha256:30674f2b5f6351aa09702a9c8c364f6a01c27aae0c1366ae8016160d1efc56b2 \
+    --hash=sha256:a0461110b7865f9a271aa1b51e516c9a95de9d696734a2f71e3e78f46e1d4678
+    # via specify-cli (pyproject.toml)
+pyyaml==6.0.3 \
+    --hash=sha256:00c4bdeba853cc34e7dd471f16b4114f4162dc03e6b7afcc2128711f0eca823c \
+    --hash=sha256:0150219816b6a1fa26fb4699fb7daa9caf09eb1999f3b70fb6e786805e80375a \
+    --hash=sha256:02893d100e99e03eda1c8fd5c441d8c60103fd175728e23e431db1b589cf5ab3 \
+    --hash=sha256:02ea2dfa234451bbb8772601d7b8e426c2bfa197136796224e50e35a78777956 \
+    --hash=sha256:0f29edc409a6392443abf94b9cf89ce99889a1dd5376d94316ae5145dfedd5d6 \
+    --hash=sha256:10892704fc220243f5305762e276552a0395f7beb4dbf9b14ec8fd43b57f126c \
+    --hash=sha256:16249ee61e95f858e83976573de0f5b2893b3677ba71c9dd36b9cf8be9ac6d65 \
+    --hash=sha256:1d37d57ad971609cf3c53ba6a7e365e40660e3be0e5175fa9f2365a379d6095a \
+    --hash=sha256:1ebe39cb5fc479422b83de611d14e2c0d3bb2a18bbcb01f229ab3cfbd8fee7a0 \
+    --hash=sha256:214ed4befebe12df36bcc8bc2b64b396ca31be9304b8f59e25c11cf94a4c033b \
+    --hash=sha256:2283a07e2c21a2aa78d9c4442724ec1eb15f5e42a723b99cb3d822d48f5f7ad1 \
+    --hash=sha256:22ba7cfcad58ef3ecddc7ed1db3409af68d023b7f940da23c6c2a1890976eda6 \
+    --hash=sha256:27c0abcb4a5dac13684a37f76e701e054692a9b2d3064b70f5e4eb54810553d7 \
+    --hash=sha256:28c8d926f98f432f88adc23edf2e6d4921ac26fb084b028c733d01868d19007e \
+    --hash=sha256:2e71d11abed7344e42a8849600193d15b6def118602c4c176f748e4583246007 \
+    --hash=sha256:34d5fcd24b8445fadc33f9cf348c1047101756fd760b4dacb5c3e99755703310 \
+    --hash=sha256:37503bfbfc9d2c40b344d06b2199cf0e96e97957ab1c1b546fd4f87e53e5d3e4 \
+    --hash=sha256:3c5677e12444c15717b902a5798264fa7909e41153cdf9ef7ad571b704a63dd9 \
+    --hash=sha256:3ff07ec89bae51176c0549bc4c63aa6202991da2d9a6129d7aef7f1407d3f295 \
+    --hash=sha256:41715c910c881bc081f1e8872880d3c650acf13dfa8214bad49ed4cede7c34ea \
+    --hash=sha256:418cf3f2111bc80e0933b2cd8cd04f286338bb88bdc7bc8e6dd775ebde60b5e0 \
+    --hash=sha256:44edc647873928551a01e7a563d7452ccdebee747728c1080d881d68af7b997e \
+    --hash=sha256:4a2e8cebe2ff6ab7d1050ecd59c25d4c8bd7e6f400f5f82b96557ac0abafd0ac \
+    --hash=sha256:4ad1906908f2f5ae4e5a8ddfce73c320c2a1429ec52eafd27138b7f1cbe341c9 \
+    --hash=sha256:501a031947e3a9025ed4405a168e6ef5ae3126c59f90ce0cd6f2bfc477be31b7 \
+    --hash=sha256:5190d403f121660ce8d1d2c1bb2ef1bd05b5f68533fc5c2ea899bd15f4399b35 \
+    --hash=sha256:5498cd1645aa724a7c71c8f378eb29ebe23da2fc0d7a08071d89469bf1d2defb \
+    --hash=sha256:5cf4e27da7e3fbed4d6c3d8e797387aaad68102272f8f9752883bc32d61cb87b \
+    --hash=sha256:5e0b74767e5f8c593e8c9b5912019159ed0533c70051e9cce3e8b6aa699fcd69 \
+    --hash=sha256:5ed875a24292240029e4483f9d4a4b8a1ae08843b9c54f43fcc11e404532a8a5 \
+    --hash=sha256:5fcd34e47f6e0b794d17de1b4ff496c00986e1c83f7ab2fb8fcfe9616ff7477b \
+    --hash=sha256:5fdec68f91a0c6739b380c83b951e2c72ac0197ace422360e6d5a959d8d97b2c \
+    --hash=sha256:6344df0d5755a2c9a276d4473ae6b90647e216ab4757f8426893b5dd2ac3f369 \
+    --hash=sha256:64386e5e707d03a7e172c0701abfb7e10f0fb753ee1d773128192742712a98fd \
+    --hash=sha256:652cb6edd41e718550aad172851962662ff2681490a8a711af6a4d288dd96824 \
+    --hash=sha256:66291b10affd76d76f54fad28e22e51719ef9ba22b29e1d7d03d6777a9174198 \
+    --hash=sha256:66e1674c3ef6f541c35191caae2d429b967b99e02040f5ba928632d9a7f0f065 \
+    --hash=sha256:6adc77889b628398debc7b65c073bcb99c4a0237b248cacaf3fe8a557563ef6c \
+    --hash=sha256:79005a0d97d5ddabfeeea4cf676af11e647e41d81c9a7722a193022accdb6b7c \
+    --hash=sha256:7c6610def4f163542a622a73fb39f534f8c101d690126992300bf3207eab9764 \
+    --hash=sha256:7f047e29dcae44602496db43be01ad42fc6f1cc0d8cd6c83d342306c32270196 \
+    --hash=sha256:8098f252adfa6c80ab48096053f512f2321f0b998f98150cea9bd23d83e1467b \
+    --hash=sha256:850774a7879607d3a6f50d36d04f00ee69e7fc816450e5f7e58d7f17f1ae5c00 \
+    --hash=sha256:8d1fab6bb153a416f9aeb4b8763bc0f22a5586065f86f7664fc23339fc1c1fac \
+    --hash=sha256:8da9669d359f02c0b91ccc01cac4a67f16afec0dac22c2ad09f46bee0697eba8 \
+    --hash=sha256:8dc52c23056b9ddd46818a57b78404882310fb473d63f17b07d5c40421e47f8e \
+    --hash=sha256:9149cad251584d5fb4981be1ecde53a1ca46c891a79788c0df828d2f166bda28 \
+    --hash=sha256:93dda82c9c22deb0a405ea4dc5f2d0cda384168e466364dec6255b293923b2f3 \
+    --hash=sha256:96b533f0e99f6579b3d4d4995707cf36df9100d67e0c8303a0c55b27b5f99bc5 \
+    --hash=sha256:9c57bb8c96f6d1808c030b1687b9b5fb476abaa47f0db9c0101f5e9f394e97f4 \
+    --hash=sha256:9c7708761fccb9397fe64bbc0395abcae8c4bf7b0eac081e12b809bf47700d0b \
+    --hash=sha256:9f3bfb4965eb874431221a3ff3fdcddc7e74e3b07799e0e84ca4a0f867d449bf \
+    --hash=sha256:a33284e20b78bd4a18c8c2282d549d10bc8408a2a7ff57653c0cf0b9be0afce5 \
+    --hash=sha256:a80cb027f6b349846a3bf6d73b5e95e782175e52f22108cfa17876aaeff93702 \
+    --hash=sha256:b30236e45cf30d2b8e7b3e85881719e98507abed1011bf463a8fa23e9c3e98a8 \
+    --hash=sha256:b3bc83488de33889877a0f2543ade9f70c67d66d9ebb4ac959502e12de895788 \
+    --hash=sha256:b865addae83924361678b652338317d1bd7e79b1f4596f96b96c77a5a34b34da \
+    --hash=sha256:b8bb0864c5a28024fac8a632c443c87c5aa6f215c0b126c449ae1a150412f31d \
+    --hash=sha256:ba1cc08a7ccde2d2ec775841541641e4548226580ab850948cbfda66a1befcdc \
+    --hash=sha256:bdb2c67c6c1390b63c6ff89f210c8fd09d9a1217a465701eac7316313c915e4c \
+    --hash=sha256:c1ff362665ae507275af2853520967820d9124984e0f7466736aea23d8611fba \
+    --hash=sha256:c2514fceb77bc5e7a2f7adfaa1feb2fb311607c9cb518dbc378688ec73d8292f \
+    --hash=sha256:c3355370a2c156cffb25e876646f149d5d68f5e0a3ce86a5084dd0b64a994917 \
+    --hash=sha256:c458b6d084f9b935061bc36216e8a69a7e293a2f1e68bf956dcd9e6cbcd143f5 \
+    --hash=sha256:d0eae10f8159e8fdad514efdc92d74fd8d682c933a6dd088030f3834bc8e6b26 \
+    --hash=sha256:d76623373421df22fb4cf8817020cbb7ef15c725b9d5e45f17e189bfc384190f \
+    --hash=sha256:ebc55a14a21cb14062aa4162f906cd962b28e2e9ea38f9b4391244cd8de4ae0b \
+    --hash=sha256:eda16858a3cab07b80edaf74336ece1f986ba330fdb8ee0d6c0d68fe82bc96be \
+    --hash=sha256:ee2922902c45ae8ccada2c5b501ab86c36525b883eff4255313a253a3160861c \
+    --hash=sha256:efd7b85f94a6f21e4932043973a7ba2613b059c4a000551892ac9f1d11f5baf3 \
+    --hash=sha256:f7057c9a337546edc7973c0d3ba84ddcdf0daa14533c2065749c9075001090e6 \
+    --hash=sha256:fa160448684b4e94d80416c0fa4aac48967a969efe22931448d853ada8baf926 \
+    --hash=sha256:fc09d0aa354569bc501d4e787133afc08552722d3ab34836a80547331bb5d4a0
+    # via specify-cli (pyproject.toml)
+readchar==4.2.2 \
+    --hash=sha256:92daf7e42c52b0787e6c75d01ecfb9a94f4ceff3764958b570c1dddedd47b200 \
+    --hash=sha256:e3b270fe16fc90c50ac79107700330a133dd4c63d22939f5b03b4f24564d5dd8
+    # via specify-cli (pyproject.toml)
+rich==15.0.0 \
+    --hash=sha256:33bd4ef74232fb73fe9279a257718407f169c09b78a87ad3d296f548e27de0bb \
+    --hash=sha256:edd07a4824c6b40189fb7ac9bc4c52536e9780fbbfbddf6f1e2502c31b068c36
+    # via
+    #   specify-cli (pyproject.toml)
+    #   typer
+shellingham==1.5.4 \
+    --hash=sha256:7ecfff8f2fd72616f7481040475a65b2bf8af90a56c89140852d1120324e8686 \
+    --hash=sha256:8dbca0739d487e5bd35ab3ca4b36e11c4078f3a234bfce294b0a0291363404de
+    # via typer
+typer==0.27.0 \
+    --hash=sha256:629bd12ea5d13a17148125d9a264f949eb171fb3f120f9b04d85873cab054fa5 \
+    --hash=sha256:6f4b27631e47f077871b7dc30e933ec0131c1390fbe0e387ea5574b5bac9ccf1
+    # via specify-cli (pyproject.toml)

.github/workflows/add-community-bundle.md

@@ -0,0 +1,288 @@
+---
+description: "Process community bundle submission issues - validate, add to catalog, and open a PR for maintainer review"
+emoji: "📦"
+
+on:
+  issues:
+    types: [labeled]
+    names: [bundle-submission]
+  skip-bots: [github-actions, copilot, dependabot]
+
+tools:
+  edit:
+  bash: ["echo", "grep", "sort", "python3", "jq", "date"]
+  github:
+    toolsets: [issues, repos]
+    min-integrity: none
+  web-fetch:
+
+permissions:
+  contents: read
+  issues: read
+
+checkout:
+  fetch-depth: 0
+
+safe-outputs:
+  noop:
+    report-as-issue: false
+  create-pull-request:
+    title-prefix: "[bundle] "
+    labels: [bundle-submission, automated]
+    draft: true
+    max: 1
+    allowed-files:
+      - bundles/catalog.community.json
+      - docs/community/bundles.md
+    protected-files:
+      policy: blocked
+      exclude:
+        - README.md
+        - CHANGELOG.md
+  add-comment:
+    max: 2
+  add-labels:
+    allowed: [bundle-submission, validation-passed, validation-failed, needs-info]
+    max: 3
+  remove-labels:
+    allowed: [validation-passed, validation-failed, needs-info]
+---
+
+# Add Community Bundle from Issue Submission
+
+You are a catalog maintenance agent for the Spec Kit project. Process community
+bundle submission issues and create draft pull requests that add or update
+entries in the community bundle catalog.
+
+Community bundles are untrusted. Validate metadata and distribution evidence,
+but do not claim to audit, endorse, or support bundle code or the components it
+installs. Never register a submitted companion catalog automatically.
+
+## Triggering Conditions
+
+This workflow is triggered by an `issues: labeled` event and is gated to the
+`bundle-submission` label. Before processing, verify that the issue title starts
+with `[Bundle]:`. If it does not, stop without commenting.
+
+## Step 1 - Read and Parse the Issue
+
+Read issue #${{ github.event.issue.number }} and extract these issue-form fields:
+
+| Field | Issue Form ID | Required |
+|-------|---------------|----------|
+| Bundle ID | `bundle-id` | Yes |
+| Bundle Name | `bundle-name` | Yes |
+| Version | `version` | Yes |
+| Role or Team | `role` | Yes |
+| Description | `description` | Yes |
+| Author | `author` | Yes |
+| Repository URL | `repository` | Yes |
+| Download URL | `download-url` | Yes |
+| Documentation URL | `documentation` | Yes |
+| License | `license` | Yes |
+| Required Spec Kit Version | `speckit-version` | Yes |
+| Integration Target | `integration` | No |
+| Components Provided | `components-provided` | Yes |
+| Required Component Catalogs | `required-catalogs` | Yes |
+| Tags | `tags` | Yes |
+| Key Features | `features` | Yes |
+| Testing Details | `testing-details` | Yes |
+| Example Usage | `example-usage` | Yes |
+| Proposed Catalog Entry | `catalog-entry` | Yes |
+
+Issue-form values appear beneath headings matching their labels.
+
+## Step 2 - Validate the Submission
+
+Run every check and collect all failures before deciding the outcome.
+
+### 2a. Bundle ID and version
+
+- The bundle ID must match
+  `^[a-z0-9](?:[a-z0-9._-]*[a-z0-9])?$`.
+- The version must be semantic version `X.Y.Z` with digits only and no `v`
+  prefix.
+
+### 2b. Repository and documentation
+
+- Restrict repository and documentation URLs to public GitHub URLs before
+  fetching them.
+- Confirm the repository exists and contains `bundle.yml`, `README.md`, and a
+  license file (`LICENSE`, `LICENSE.md`, or `LICENSE.txt`).
+- The documentation URL must resolve to a readable Markdown file that explains
+  the bundle's intended role, installed components, required catalogs, and
+  installation steps.
+- Confirm the repository's `bundle.yml` matches the submitted bundle ID,
+  version, role, author, license, Spec Kit requirement, integration target, and
+  component summary.
+
+### 2c. Release artifact
+
+- The download URL must be an HTTPS GitHub release asset URL under the submitted
+  repository:
+  `https://github.com/<owner>/<repo>/releases/download/<tag>/<asset>.zip`.
+- Confirm the release exists, its tag corresponds to the submitted version
+  (`vX.Y.Z` or `X.Y.Z`), and the exact ZIP asset is attached to that release.
+- Confirm the asset name is versioned and consistent with the submitted bundle
+  ID and version.
+
+Do not fetch arbitrary user-provided URLs. Do not claim the artifact was
+executed or audited; rely on the required submission attestations for build and
+installation evidence.
+
+### 2d. Catalog entry
+
+Parse the proposed JSON and require one entry under the submitted bundle ID.
+Confirm that:
+
+- `id`, `name`, `version`, `role`, `description`, `author`, `license`,
+  `download_url`, and `repository` match the submission and manifest.
+- `requires.speckit_version` matches the submission.
+- `provides` contains non-negative integer counts for `extensions`, `presets`,
+  `steps`, and `workflows`, matching the manifest.
+- `tags` contains 2-5 lowercase strings and matches the submitted tags.
+- `verified` is the boolean value `false`. Community entries must never be
+  marked verified.
+
+### 2e. Component resolution
+
+- `Required Component Catalogs` must explicitly say `None` or list every
+  non-default extension, preset, workflow, and step catalog needed by the
+  bundle.
+- Compare the manifest references, README, required-catalog field, testing
+  details, and example usage for consistency.
+- If non-default catalogs are required, ensure each URL is HTTPS, the README
+  documents the corresponding `catalog add` command, and the testing details
+  say those catalogs were registered in the clean-project test.
+- If the field says `None` but a component is not bundled and cannot be
+  installed from a default Spec Kit catalog, fail validation and ask the
+  submitter to list and document an install-allowed companion catalog.
+
+The community bundle catalog itself remains discovery-only. Companion catalog
+URLs are documentation and validation metadata, not catalogs this workflow
+should add to Spec Kit.
+
+### 2f. Checklists and testing evidence
+
+- Confirm every required checkbox in Testing Checklist and Submission
+  Requirements is checked (`[x]`).
+- Confirm Testing Details describe validation, build, artifact installation,
+  and clean-project testing.
+- Confirm Example Usage includes artifact installation and, when applicable,
+  all required catalog setup commands.
+
+### Validation outcome
+
+If any check fails:
+
+1. Comment once with every failed check and a specific correction.
+2. Remove `validation-passed`.
+3. Add `validation-failed`; add `needs-info` when submitter input is needed.
+4. Stop without editing files or creating a pull request.
+
+If all checks pass, remove `validation-failed` and `needs-info`, add
+`validation-passed`, and continue.
+
+## Step 3 - Determine Add or Update
+
+Search `bundles/catalog.community.json` for the bundle ID.
+
+- If absent, add a new entry.
+- If present, update the existing entry in place.
+
+Treat a submitted version lower than or equal to the existing catalog version
+as a validation failure unless the issue clearly documents a metadata-only
+correction at the same version.
+
+## Step 4 - Update the Community Catalog
+
+Edit `bundles/catalog.community.json`. Insert new entries alphabetically by
+bundle ID. The entry shape is:
+
+```json
+{
+  "<bundle-id>": {
+    "name": "<bundle-name>",
+    "id": "<bundle-id>",
+    "version": "<version>",
+    "role": "<role>",
+    "description": "<description>",
+    "author": "<author>",
+    "license": "<license>",
+    "download_url": "<download-url>",
+    "repository": "<repository>",
+    "requires": {
+      "speckit_version": "<speckit-version>"
+    },
+    "provides": {
+      "extensions": 0,
+      "presets": 0,
+      "steps": 0,
+      "workflows": 0
+    },
+    "tags": ["<tag>"],
+    "verified": false
+  }
+}
+```
+
+Use the validated proposed entry rather than inventing metadata. Keep
+`verified: false`. Update the top-level `updated_at` to today's UTC date at
+midnight and preserve the top-level `catalog_url`.
+
+Validate the complete file:
+
+```bash
+python3 -c "import json; json.load(open('bundles/catalog.community.json')); print('Valid JSON')"
+```
+
+## Step 5 - Update Community Documentation
+
+Add or update the bundle in `docs/community/bundles.md`. Keep rows alphabetical
+by bundle name:
+
+```text
+| <Name> | <Description> | `<role>` | <component counts> | <None or documented> | [<repo-name>](<repository>) |
+```
+
+Before rendering the row, convert every user-derived display value to
+single-line plain text: collapse CR/LF sequences to spaces, remove control
+characters, and backslash-escape `\`, `|`, backticks, `*`, `_`, `[`, `]`, `<`,
+and `>`. Use the validated HTTPS GitHub repository URL unchanged only as the
+Markdown link destination.
+
+Render component counts compactly, omitting zero-valued component types. Use
+`None` when no companion catalogs are needed and `Documented` otherwise; the
+repository README remains the source for the actual URLs.
+
+## Step 6 - Create a Draft Pull Request
+
+Create one draft pull request.
+
+- New entry branch:
+  `community/${{ github.event.issue.number }}-add-<bundle-id>-bundle`
+- Update branch:
+  `community/${{ github.event.issue.number }}-update-<bundle-id>-bundle`
+- New title: `Add <Bundle Name> bundle to community catalog`
+- Update title: `Update <Bundle Name> bundle to v<version>`
+
+The commit and PR description must summarize the catalog and documentation
+changes, list the validation results, include
+`Closes #${{ github.event.issue.number }}`, and mention the submitter with
+`cc @<issue-author>`.
+
+End the commit message with this authorship trailer:
+
+```text
+Assisted-by: GitHub Copilot (model: <name-if-known>, autonomous)
+```
+
+## Important Rules
+
+- Modify only `bundles/catalog.community.json` and
+  `docs/community/bundles.md`.
+- Keep JSON entries sorted by ID and documentation rows sorted by name.
+- Never set a community bundle's `verified` field to true.
+- Never add, enable, or change the policy of a submitted catalog.
+- Never describe validation as a security audit or endorsement.
+- Use `Closes`, not `Fixes`, for the submission issue.

.github/workflows/catalog-assign.yml

@@ -9,11 +9,13 @@ jobs:
     if: >
       (github.event.action == 'opened' && (
         contains(github.event.issue.labels.*.name, 'extension-submission') ||
-        contains(github.event.issue.labels.*.name, 'preset-submission')
+        contains(github.event.issue.labels.*.name, 'preset-submission') ||
+        contains(github.event.issue.labels.*.name, 'bundle-submission')
       )) ||
       (github.event.action == 'labeled' && (
         github.event.label.name == 'extension-submission' ||
-        github.event.label.name == 'preset-submission'
+        github.event.label.name == 'preset-submission' ||
+        github.event.label.name == 'bundle-submission'
       ))
     runs-on: ubuntu-latest
     permissions:

.github/workflows/security.yml

@@ -0,0 +1,78 @@
+name: Security Audit
+
+permissions:
+  contents: read
+
+on:
+  push:
+    branches: ["main"]
+  pull_request:
+    types: [opened, synchronize, reopened]
+  schedule:
+    - cron: "17 4 * * 1"
+  workflow_dispatch:
+
+jobs:
+  dependency-audit:
+    name: Dependency audit
+    if: ${{ github.event_name != 'schedule' }}
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
+        with:
+          fetch-depth: 0
+
+      - name: Install uv
+        uses: astral-sh/setup-uv@11f9893b081a58869d3b5fccaea48c9e9e46f990 # v8.3.2
+
+      - name: Set up Python
+        uses: actions/setup-python@ece7cb06caefa5fff74198d8649806c4678c61a1 # v6
+        with:
+          python-version: "3.14"
+
+      - name: Check committed audit requirements are current
+        env:
+          DEPENDENCY_DIFF_BASE: ${{ github.event.pull_request.base.sha || github.event.before || '' }}
+          DEPENDENCY_DIFF_HEAD: ${{ github.sha }}
+          GENERATED_REQUIREMENTS: ${{ runner.temp }}/security-audit-requirements.txt
+        run: python .github/scripts/check_security_requirements.py
+
+      - name: Run pip-audit (committed requirements)
+        run: uvx --from pip-audit==2.10.0 pip-audit --disable-pip --require-hashes -r .github/security-audit-requirements.txt --progress-spinner off
+
+  dependency-audit-scheduled:
+    name: Dependency audit scheduled (${{ matrix.os }}, Python ${{ matrix.python-version }})
+    if: ${{ github.event_name == 'schedule' }}
+    runs-on: ${{ matrix.os }}
+    strategy:
+      fail-fast: false
+      matrix:
+        os: [ubuntu-latest, windows-latest]
+        python-version: ["3.11", "3.12", "3.13", "3.14"]
+    steps:
+      - name: Checkout
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
+
+      - name: Install uv
+        uses: astral-sh/setup-uv@11f9893b081a58869d3b5fccaea48c9e9e46f990 # v8.3.2
+
+      - name: Set up Python ${{ matrix.python-version }}
+        uses: actions/setup-python@ece7cb06caefa5fff74198d8649806c4678c61a1 # v6
+        with:
+          python-version: ${{ matrix.python-version }}
+
+      # The committed .github/security-audit-requirements.txt is generated with
+      # --universal (resolves across all interpreters/platforms) and is what
+      # push/PR/workflow_dispatch runs audit. The scheduled job instead compiles
+      # per matrix entry with --python-version so it can surface advisories in
+      # wheels that only resolve on a specific interpreter (e.g. 3.11-only) —
+      # coverage the universal file may not exercise. This broadening is
+      # intentional; non-scheduled runs trade that depth for determinism against
+      # the committed snapshot.
+      - name: Compile scheduled audit requirements
+        run: |
+          uv pip compile pyproject.toml --extra test --python-version "${{ matrix.python-version }}" --upgrade --generate-hashes --quiet --output-file "${{ runner.temp }}/spec-kit-audit-requirements.txt"
+
+      - name: Run pip-audit (scheduled live resolution)
+        run: uvx --from pip-audit==2.10.0 pip-audit --disable-pip --require-hashes -r "${{ runner.temp }}/spec-kit-audit-requirements.txt" --progress-spinner off

.github/workflows/test.yml

@@ -24,7 +24,7 @@ jobs:
           python-version: "3.14"
 
       - name: Run ruff check
-        run: uvx ruff check src/
+        run: uvx ruff check src tests
 
   pytest:
     runs-on: ${{ matrix.os }}

CHANGELOG.md

@@ -2,6 +2,101 @@
 
 <!-- insert new changelog below this comment -->
 
+## [0.13.4] - 2026-07-22
+
+### Changed
+
+- docs(concepts): document the spec-of-specs feature breakdown approach (#3648)
+- fix(scripts): git-ext PowerShell emits the '# To persist' SPECIFY_FEATURE hint (parity) (#3632)
+- fix(integrations): validate cached catalog shape before returning it (#3627)
+- fix(bundler): reject non-list 'catalogs' in bundle-catalogs.yml with a clean error (#3623)
+- fix(bundler): guard lazy .hostname ValueError in catalog add_source (#3644)
+- Add Intake Authoring Governance preset to community catalog (#3643)
+- feat: add Factory Droid CLI integration (#822) (#3587)
+- docs(installation): document the 'py' (Python) script type (#3640)
+- fix(init): show hyphenated /speckit-<name> in Next Steps for Forge projects (#3642)
+- fix(extensions): render hyphenated hook invocations for Forge projects (#3641)
+- fix(workflows): workflow add detects local YAML files case-insensitively (#3633)
+- fix(workflows): list-literal expression ignores trailing/empty commas (#3631)
+- fix(workflows): StepRegistry.add tolerates a corrupted non-dict existing entry (#3630)
+- fix(bundler): reject non-mapping 'integration' in a bundle manifest (#3629)
+- fix(workflows): command/prompt steps fail cleanly on a non-string integration (#3626)
+- docs(core): document the 'py' (Python) --script type in the init option table (#3625)
+- fix(workflows): gate prompt uses isdecimal() so a superscript digit doesn't crash (#3624)
+- fix(integrations): Cline dispatches hyphenated /speckit-<cmd> invocations (#3622)
+- docs(upgrade): document integration upgrade / extension update as the project-files upgrade path (#3326)
+- chore: release 0.13.3, begin 0.13.4.dev0 development (#3645)
+
+## [0.13.3] - 2026-07-22
+
+### Changed
+
+- fix(integrations): escape Rich markup in --integration-options error messages (#3458)
+- docs: document __SPECKIT_COMMAND_ token for portable cross-command references (#3503)
+- [preset] Add Parallel Autonomous Run Governance preset to community catalog (#3614)
+- docs(workflows): fix stale FanOutStep docstring claiming sequential-only execution (#3639)
+- [bundle] Add SicarioSpec Security & Governance Bundle to community catalog (#3636)
+- [preset] Update Autonomous Run Governance preset to v0.3.2 (#3615)
+- fix(workflows): validate every redirect hop when fetching workflow/step catalogs (#3637)
+- Add pipeline workflow to community catalog (#3338)
+- [extension] Add Linear Weave extension to community catalog (#3609)
+- docs: clarify hook priority validation semantics (#3594)
+- fix(workflows): reject a non-string 'integration'/'model' in command & prompt steps (#3597)
+- ci: add dependency audit workflow (#3138)
+- Add Intake Review Governance preset to community catalog (#3613)
+- fix(workflows): reject non-list input 'enum' instead of crashing (#3601)
+- chore: release 0.13.2, begin 0.13.3.dev0 development (#3617)
+
+## [0.13.2] - 2026-07-21
+
+### Changed
+
+- fix(workflows): reject a non-string 'command' in command-step (#3596)
+- fix(workflows): fail gate step loudly on a malformed 'options' (#3595)
+- fix(extensions): re-validate catalog URL after redirects (HTTPS parity/security) (#3524)
+- Add community bundle submission automation (#3553)
+- fix(presets): re-validate catalog URL after redirects (HTTPS parity/security) (#3523)
+- feat(scripts): port create-new-feature, setup-plan and setup-tasks to Python (#3386)
+- fix(agents): parse frontmatter on the --- delimiter line, not any --- substring (#3590)
+- [bug-fix] Fix reinstall-overwrites-kept-config: preserve config on plain reinstall after --keep-config (#3449)
+- feat: update Bob integration to skills-based layout for Bob 2.0 (#3415)
+- Update OKF Knowledge Bundle Generator to v0.3.0 (#3608)
+- Add Test Coverage Drift Control extension to community catalog (#3607)
+- chore: align ruff lint scope (#3139)
+- feat(workflows): WorkflowResolver standalone (PR 1) (#3557)
+- fix(extensions,presets): surface clean error on malformed download URL (#3577)
+- chore: release 0.13.1, begin 0.13.2.dev0 development (#3610)
+
+## [0.13.1] - 2026-07-21
+
+### Changed
+
+- fix(integrations): catch OverflowError on a `priority: .inf` in add/remove (#3589)
+- fix(workflows): reject bool / .inf catalog priority in workflow & step catalog loaders (#3526)
+- fix(catalogs): 'priority: .inf' yields a clean validation error instead of crashing (#3525)
+- docs(integrations): document the 'integration list --catalog' flag (#3530)
+- fix(workflows): fail fan-in loudly on a non-string wait_for entry (#3579)
+- fix(workflows): fail fan-out loudly on a truthy non-mapping step template (#3537)
+- fix(workflows): reject a non-string prompt in prompt-step validate() (#3582)
+- fix(workflows): route 'workflow status --json' errors to stderr (#3520)
+- fix(integrations): Forge dispatches hyphenated /speckit-<cmd> invocations (#3529)
+- chore: release 0.13.0, begin 0.13.1.dev0 development (#3588)
+
+## [0.13.0] - 2026-07-17
+
+### Changed
+
+- fix(auth): Azure DevOps az-CLI token acquisition returns None on undecodable output (#3527)
+- feat(extensions): add assess idea assessment pipeline extension (#3568)
+- fix(bundle): surface a clean BundlerError on a malformed bundle download URL (#3586)
+- Add OKF Knowledge Bundle Generator extension to community catalog (#3585)
+- Update Autonomous Run Governance preset to v0.2.2 (#3584)
+- docs: update extension guide PyPI upgrade guidance (#3578)
+- fix(presets): raise PresetValidationError, not raw ValueError, on malformed catalog URL (#3576)
+- chore(deps): bump github/codeql-action/init from 4.36.2 to 4.37.1 (#3571)
+- docs: align README hero tagline and subtitle with docs/index.md (#3581)
+- chore: release 0.12.18, begin 0.12.19.dev0 development (#3583)
+
 ## [0.12.18] - 2026-07-17
 
 ### Changed

CONTRIBUTING.md

@@ -113,6 +113,27 @@ uv pip install -e ".[test]"
 > `specify_cli` to this checkout's `src/`. This matches the gotcha documented in
 > `AGENTS.md` (Common Pitfalls).
 
+#### Security checks
+
+```bash
+uvx --from pip-audit==2.10.0 pip-audit --disable-pip --require-hashes -r .github/security-audit-requirements.txt --progress-spinner off
+```
+
+This command audits the committed hashed requirements snapshot. Pull request,
+push, and manual CI runs use the same snapshot so their results stay
+deterministic. If dependency metadata changes, refresh and commit the snapshot
+before auditing it:
+
+```bash
+uv pip compile pyproject.toml --extra test --universal --upgrade --generate-hashes --quiet --no-header --output-file .github/security-audit-requirements.txt
+```
+
+The scheduled CI audit resolves the runtime and `test` extra dependency set
+across the supported Python and OS matrix to catch newly published advisories.
+Upstream package releases drift over time, so even an unrelated PR touching
+`pyproject.toml` can fail the `dependency-audit` check until the committed file
+is regenerated with the command above and re-committed.
+
 #### Shell scripts
 
 ```bash

bundles/catalog.community.json

@@ -0,0 +1,35 @@
+{
+  "schema_version": "1.0",
+  "updated_at": "2026-07-22T00:00:00Z",
+  "catalog_url": "https://raw.githubusercontent.com/github/spec-kit/main/bundles/catalog.community.json",
+  "bundles": {
+    "sicario-spec": {
+      "name": "SicarioSpec Security & Governance Bundle",
+      "id": "sicario-spec",
+      "version": "0.5.1",
+      "role": "security-engineer",
+      "description": "Secure-by-default governance bundle for GitHub Spec Kit. Enforces data classification, threat modeling, and code-owned verification gates.",
+      "author": "SicarioSpec Contributors",
+      "license": "MIT",
+      "download_url": "https://github.com/dfirs1car1o/sicario-spec/releases/download/v0.5.1/sicario-spec-0.5.1.zip",
+      "repository": "https://github.com/dfirs1car1o/sicario-spec",
+      "requires": {
+        "speckit_version": ">=0.9.0"
+      },
+      "provides": {
+        "extensions": 1,
+        "presets": 11,
+        "steps": 0,
+        "workflows": 0
+      },
+      "tags": [
+        "security",
+        "governance",
+        "compliance",
+        "appsec",
+        "threat-modeling"
+      ],
+      "verified": false
+    }
+  }
+}

docs/community/bundles.md

@@ -5,7 +5,11 @@
 
 Bundles compose existing Spec Kit components — extensions, presets, workflows, and steps — into a single role or team stack. They are useful when a user should be able to install a tested set of components together instead of following several separate install commands.
 
-Accepted community bundle entries will be listed here once a community bundle catalog is available. To submit a bundle for review, file a [Bundle Submission](https://github.com/github/spec-kit/issues/new?template=bundle_submission.yml) issue.
+Accepted community bundle entries are published in [`bundles/catalog.community.json`](https://github.com/github/spec-kit/blob/main/bundles/catalog.community.json) and listed below. The built-in community source is discovery-only: `specify bundle search` and `specify bundle info` can inspect entries, but installing by ID requires explicitly adding an install-allowed catalog. Explicit catalogs use a higher default precedence than the built-in community source. To submit a bundle for review, file a [Bundle Submission](https://github.com/github/spec-kit/issues/new?template=bundle_submission.yml) issue.
+
+| Bundle | Purpose | Role or team | Provides | Required catalogs | URL |
+|--------|---------|--------------|----------|-------------------|-----|
+| SicarioSpec Security & Governance Bundle | Secure-by-default governance bundle for GitHub Spec Kit. Enforces data classification, threat modeling, and code-owned verification gates. | `security-engineer` | 1 extension, 11 presets | Documented | [sicario-spec](https://github.com/dfirs1car1o/sicario-spec) |
 
 ## What to Submit
 

docs/community/extensions.md

@@ -70,6 +70,7 @@ The following community-contributed extensions are available in [`catalog.commun
 | Jira Integration (Sync Engine) | Idempotent, drift-aware, fail-closed reconcile engine mirroring spec-kit specs into Jira (Epic per repo, Story per spec, Subtask per phase) | `integration` | Read+Write | [spec-kit-jira-sync](https://github.com/ashbrener/spec-kit-jira-sync) |
 | Learning Extension | Generate educational guides from implementations and enhance clarifications with mentoring context | `docs` | Read+Write | [spec-kit-learn](https://github.com/imviancagrace/spec-kit-learn) |
 | Linear Integration | Mirror spec-kit feature directories into Linear (filesystem → Linear, reconcile-based, unidirectional). | `integration` | Read+Write | [spec-kit-linear-sync](https://github.com/ashbrener/spec-kit-linear-sync) |
+| Linear Weave | Weave Spec Kit into Linear: pull requirements, mirror tasks.md into sub-issues, sync statuses | `integration` | Read+Write | [spec-kit-linear-weave](https://github.com/tonydwoodhouse/spec-kit-linear-weave) |
 | LLM Wiki | LLM-maintained compounding project wiki: source ingestion, cited answers, and consistency linting | `docs` | Read+Write | [spec-kit-wiki](https://github.com/formin/spec-kit-wiki) |
 | Loop Engineering | Engineer safe autonomous agent loops for spec-driven development: a maker/checker split, externalized loop state, and stay-the-engineer guardrails against comprehension debt and cognitive surrender | `process` | Read+Write | [spec-kit-loop](https://github.com/formin/spec-kit-loop) |
 | MAQA — Multi-Agent & Quality Assurance | Coordinator → feature → QA agent workflow with parallel worktree-based implementation. Language-agnostic. Auto-detects installed board plugins. Optional CI gate. | `process` | Read+Write | [spec-kit-maqa-ext](https://github.com/GenieRobot/spec-kit-maqa-ext) |
@@ -89,7 +90,7 @@ The following community-contributed extensions are available in [`catalog.commun
 | Multi-Repo Branch Sync | Creates the feature branch in affected sub-repositories and git submodules via plan/tasks hooks | `process` | Read+Write | [multi-repo-sync](https://github.com/fyloss/spec-kit-multi-repo-sync) |
 | Multi-Sites Spec Kit | Multi-site aware specify command with per-site spec folders, auto-increment, and Drupal support | `process` | Read+Write | [spec-kit-multi-sites](https://github.com/teeyo/spec-kit-multi-sites) |
 | .NET Framework to Modern .NET Migration | Orchestrate end-to-end .NET Framework to modern .NET migration across 7 phases, with SDD lifecycle integration | `process` | Read+Write | [spec-kit-fx-to-net](https://github.com/RogerBestMsft/spec-kit-FxToNet) |
-| OKF Knowledge Bundle Generator | Generates and maintains an Open Knowledge Format (OKF v0.1) knowledge bundle from a source-code repository | `docs` | Read+Write | [speckit_ofk](https://github.com/alexcpn/speckit_ofk) |
+| OKF Knowledge Bundle Generator | Generates and maintains an Open Knowledge Format (OKF v0.1) knowledge bundle from a source-code repository, mining git history for significance and rationale, and resolving open questions with the user | `docs` | Read+Write | [speckit_ofk](https://github.com/alexcpn/speckit_ofk) |
 | Onboard | Contextual onboarding and progressive growth for developers new to spec-kit projects. Explains specs, maps dependencies, validates understanding, and guides the next step | `process` | Read+Write | [spec-kit-onboard](https://github.com/dmux/spec-kit-onboard) |
 | Optimize | Audit and optimize AI governance for context efficiency — token budgets, rule health, interpretability, compression, coherence, and echo detection | `process` | Read+Write | [spec-kit-optimize](https://github.com/sakitA/spec-kit-optimize) |
 | Orchestration Task Context Management | Adds subagent work-unit orchestration to generated Spec Kit task files | `process` | Read+Write | [spec-kit-orchestration-task-context-management](https://github.com/benizzio/spec-kit-orchestration-task-context-management) |
@@ -149,6 +150,7 @@ The following community-contributed extensions are available in [`catalog.commun
 | Superspec | Bridges spec-kit with obra/superpowers (brainstorming, TDD, subagent, code-review) into a unified, resumable workflow with graceful degradation and session progress tracking | `process` | Read+Write | [superspec](https://github.com/WangX0111/superspec) |
 | Tasks to GitHub Project | Publish and synchronize Spec Kit tasks as cards on a GitHub Project (v2) kanban board, with priority and status sync between spec.md/tasks.md and the board. | `integration` | Read+Write | [spec-kit-tasks-to-project](https://github.com/mancioshell/spec-kit-tasks-to-project) |
 | Team Assign | Assign tasks.md items to human engineers, split into subtasks, and generate a per-engineer workboard | `process` | Read+Write | [spec-kit-team-assign](https://github.com/tarunkumarbhati/spec-kit-team-assign) |
+| Test Coverage Drift Control | Generate incremental coverage drift reports and planned remediation tasks after implementation | `code` | Read+Write | [spec-kit-test-coverage-drift-control](https://github.com/benizzio/spec-kit-test-coverage-drift-control) |
 | Time Machine | Retroactively apply the full SDD workflow to existing codebases — analyse, spec, and ship feature-by-feature | `process` | Read+Write | [spec-kit-time-machine](https://github.com/teeyo/spec-kit-time-machine) |
 | TinySpec | Lightweight single-file workflow for small tasks — skip the heavy multi-step SDD process | `process` | Read+Write | [spec-kit-tinyspec](https://github.com/Quratulain-bilal/spec-kit-tinyspec) |
 | Token Budget | Reduces LLM token consumption in Spec Kit workflows: compact artifacts in-place, scope per-phase reading, suppress prose padding, and report token usage | `process` | Read+Write | [spec-kit-token-budget](https://github.com/tinesoft/spec-kit-token-budget) |

docs/community/presets.md

@@ -11,7 +11,7 @@ The following community-contributed presets customize how Spec Kit behaves — o
 | Agent Parity Governance | Adds shared-guidance parity, audit-ready Spec-Kit run evidence, and agent-neutral model-routing guidance across a project's declared AI-agent instruction surfaces so agent guidance does not drift. | 6 templates, 3 commands | — | [spec-kit-preset-agent-parity-governance](https://github.com/hindermath/spec-kit-preset-agent-parity-governance) |
 | AIDE In-Place Migration | Adapts the AIDE extension workflow for in-place technology migrations (X → Y pattern) — adds migration objectives, verification gates, knowledge documents, and behavioral equivalence criteria | 2 templates, 8 commands | AIDE extension | [spec-kit-presets](https://github.com/mnriem/spec-kit-presets) |
 | Architecture Governance | Adds secure software architecture, STRIDE+CAPEC threat modeling, arc42 security cross-cutting concepts, S-ADRs, Zero Trust applicability, OWASP SAMM governance, BSI C3A cloud autonomy, BSI C5 cloud compliance assurance, and audit-ready Spec Kit run evidence | 13 templates, 3 commands | — | [spec-kit-preset-architecture-governance](https://github.com/hindermath/spec-kit-preset-architecture-governance) |
-| Autonomous Run Governance | Adds permission-bounded, evidence-first governance for autonomous Spec Kit delivery with validated status, stop, resume, exact-head proof, closeout, and learner guidance. | 13 templates, 5 commands, 4 scripts | — | [spec-kit-preset-autonomous-run-governance](https://github.com/hindermath/spec-kit-preset-autonomous-run-governance) |
+| Autonomous Run Governance | Adds permission-bounded, evidence-first governance for complete autonomous Spec Kit delivery, including validated status, stop, explicit resume, exact-head proof, post-merge closeout, retrospective learning, and an optional policy-driven intake-review gate before feature creation. | 13 templates, 5 commands, 4 scripts | — | [spec-kit-preset-autonomous-run-governance](https://github.com/hindermath/spec-kit-preset-autonomous-run-governance) |
 | Canon Core | Adapts original Spec Kit workflow to work together with Canon extension | 2 templates, 8 commands | — | [spec-kit-canon](https://github.com/maximiliamus/spec-kit-canon) |
 | Claude AskUserQuestion | Upgrades `/speckit.clarify` and `/speckit.checklist` on Claude Code from Markdown-table prompts to the native AskUserQuestion picker, with a recommended option and reasoning on every question | 2 commands | — | [spec-kit-preset-claude-ask-questions](https://github.com/0xrafasec/spec-kit-preset-claude-ask-questions) |
 | Command Density | Compacts the nine core Spec Kit command prompts while preserving scripts, handoffs, placeholders, hook output blocks, and rule structure | 9 commands | — | [spec-kit-preset-command-density](https://github.com/Xopoko/spec-kit-preset-command-density) |
@@ -19,10 +19,13 @@ The following community-contributed presets customize how Spec Kit behaves — o
 | Explicit Task Dependencies | Adds explicit `(depends on T###)` dependency declarations and an Execution Wave DAG to tasks.md for parallel scheduling | 1 template, 1 command | — | [spec-kit-preset-explicit-task-dependencies](https://github.com/Quratulain-bilal/spec-kit-preset-explicit-task-dependencies) |
 | Fiction Book Writing | It adapts the Spec-Driven Development workflow for storytelling to create books or audiobooks (with annotations) in 12 languages: features become story elements, specs become story briefs, plans become story structures, and tasks become scene-by-scene writing tasks. Supports single and multi-POV, all major plot structure frameworks, and two style modes: an author voice sample or humanized AI prose principles. Supports interactive elements like brainstorming, interview, roleplay, and extras like statistics, cover builder, illustration builder, and bio command. Export with templates for KDP, D2D, etc. | 26 templates, 34 commands, 2 scripts | — | [speckit-preset-fiction-book-writing](https://github.com/adaumann/speckit-preset-fiction-book-writing) |
 | Game Narrative Writing | Preset for game narrative design and interactive storytelling. It adapts the Spec-Driven Development workflow for game narratives: features become story mechanics, specs become narrative briefs, plans become story maps, and tasks become dialogue and scene-writing tasks. Supports branching narratives, player agency systems, state machines, and interactive dialogue trees. | 37 templates, 34 commands, 5 scripts | — | [speckit-preset-game-narrative-writing](https://github.com/adaumann/speckit-preset-game-narrative-writing) |
+| Intake Authoring Governance | Creates traceable Spec Kit intake files and receipts from ordered text sources while preserving clarification, update, and delivery-authority boundaries. | 7 templates, 2 commands, 2 scripts | — | [spec-kit-preset-intake-authoring-governance](https://github.com/hindermath/spec-kit-preset-intake-authoring-governance) |
+| Intake Review Governance | Adds hash-bound review, repair, and status gates for single, series, and campaign intake files before interactive, autonomous, or parallel Spec Kit execution. | 8 templates, 3 commands, 2 scripts | — | [spec-kit-preset-intake-review-governance](https://github.com/hindermath/spec-kit-preset-intake-review-governance) |
 | iSAQB Architecture Governance | Adds general iSAQB/CPSA-F and arc42 software-architecture governance, including audit-ready Spec Kit run evidence for architecture goals, views, quality scenarios, ADRs, risks, and technical debt. | 13 templates, 3 commands | — | [spec-kit-preset-isaqb-architecture-governance](https://github.com/hindermath/spec-kit-preset-isaqb-architecture-governance) |
 | Jira Issue Tracking | Overrides `speckit.taskstoissues` to create Jira epics, stories, and tasks instead of GitHub Issues via Atlassian MCP tools | 1 command | — | [spec-kit-preset-jira](https://github.com/luno/spec-kit-preset-jira) |
 | Model Driven Engineering | Focuses on streamlined commands, app repository support, cross-spec support, and capability-aware project memory for model-driven engineering workflows | 6 templates, 11 commands | MDE extension | [spec-kit-preset-mde](https://github.com/AI-MDE/spec-kit-preset-mde) |
 | Multi-Repo Branching | Coordinates feature branch creation across multiple git repositories (independent repos and submodules) during plan and tasks phases | 2 commands | — | [spec-kit-preset-multi-repo-branching](https://github.com/sakitA/spec-kit-preset-multi-repo-branching) |
+| Parallel Autonomous Run Governance | Coordinates isolated autonomous Spec Kit campaigns with bounded concurrency, mixed agents, resumable consolidation, governed post-merge closeout, schema 1.2, and an optional current intake-review gate before worker scheduling. | 9 templates, 5 commands, 2 scripts | autonomous-run-governance >=0.3.2; optional: intake-review-governance >=0.1.0 | [spec-kit-preset-parallel-autonomous-run-governance](https://github.com/hindermath/spec-kit-preset-parallel-autonomous-run-governance) |
 | Pirate Speak (Full) | Transforms all Spec Kit output into pirate speak — specs become "Voyage Manifests", plans become "Battle Plans", tasks become "Crew Assignments" | 6 templates, 9 commands | — | [spec-kit-presets](https://github.com/mnriem/spec-kit-presets) |
 | Screenwriting | Spec-Driven Development for screenwriting/scriptwriting/tutorials: feature films, television (pilot, episode, limited series), and stage plays. Adapts the Spec Kit workflow to screenplay craft — slug lines, action lines, act breaks, beat sheets, and industry-standard pitch documents. Supports three-act, Save the Cat, TV pilot, network episode, cable/streaming episode, and stage-play structural frameworks. Export to Fountain, FTX, PDF | 26 templates, 32 commands, 1 script | — | [speckit-preset-screenwriting](https://github.com/adaumann/speckit-preset-screenwriting) |
 | Security Governance | Adds memory-safe-language preference, language-specific secure coding profiles, audit-ready Spec-Kit run evidence, ASVS verification, SBOM/AI-SBOM supply-chain transparency, CRA awareness, and regulatory applicability screening for NIS2, CRA, EU AI Act, and DORA | 14 templates, 3 commands | — | [spec-kit-preset-security-governance](https://github.com/hindermath/spec-kit-preset-security-governance) |

docs/concepts/complex-features.md

@@ -63,10 +63,14 @@ independently specified sub-features. Each sub-feature gets its own
 `spec.md`, `plan.md`, and `tasks.md`, and runs through its own
 specify/plan/tasks/implement cycle.
 
-This is the "spec of specs" approach: the first iteration breaks a massive
-feature into smaller, self-contained specs that can each be implemented without
-overwhelming the model. It adds the most overhead, so reserve it for features
-that are too large to handle any other way.
+This is the "spec of specs" approach: a first pass breaks a massive feature into
+smaller, self-contained specs that can each be implemented without overwhelming the
+model. It adds the most overhead, so reserve it for features that are too large to
+handle any other way.
+
+See [Spec of Specs](spec-of-specs.md) for the full procedure — how to run the
+roadmap pass, structure the roadmap artifact, link sub-specs back to it, and a worked
+example.
 
 ## Which Approach to Choose
 

docs/concepts/spec-of-specs.md

@@ -0,0 +1,171 @@
+# Spec of Specs
+
+When a feature is too large to run through a single
+`/speckit.specify` → `/speckit.plan` → `/speckit.tasks` → `/speckit.implement`
+cycle without the model losing track mid-implementation, you can break it into a
+**roadmap** of smaller, independently-specified sub-features. This is the "spec of
+specs" approach: one up-front pass decomposes a massive feature into self-contained
+specs, and each of those runs through its own specify/plan/tasks/implement cycle.
+
+> **When to reach for this.** Decomposition adds the most overhead of any strategy
+> in [Handling Complex Features](complex-features.md). Use it **only when the lighter
+> options there are insufficient** — first try limiting how many tasks run per
+> `/speckit.implement` invocation, then sub-agent delegation, then a combination.
+> Reach for a spec of specs only when even a single phase is too large to handle in
+> one run.
+
+The rest of this page describes *how* to do it with the tools you already have. No
+new commands or extensions are required.
+
+## The roadmap pass
+
+Before writing any sub-spec, do a single decomposition pass to produce a roadmap.
+Treat this as a lightweight planning conversation with your agent, not a full spec:
+
+1. **State the whole feature.** Describe the large feature (the "epic") in a
+   sentence or two so the agent has the full picture up front.
+2. **Identify independent slices.** Ask the agent to propose a small set of
+   sub-features that each deliver a coherent piece of the epic and can be specified
+   on their own. Aim for slices that are independently testable — implementing just
+   one should leave you with something demonstrable.
+3. **Draw the boundaries.** For each slice, write one line of intent and an explicit
+   scope boundary (what is in, what is deferred to a sibling slice). Sharp
+   boundaries are what keep each sub-spec small enough to fit in context.
+4. **Order by dependency.** Note which slices depend on others and sequence them so
+   prerequisites come first. Slices with no dependency on each other can be built in
+   any order. To build independent slices in parallel, use separate worktrees so each
+   run has isolated active-feature state.
+5. **Record the result as a roadmap.** Capture the slices in a durable roadmap file
+   (below) so every later sub-spec can point back to it.
+
+The roadmap is deliberately shallow: it names and orders the sub-features but does
+**not** design them. The design happens when each slice runs through its own
+`/speckit.specify`.
+
+## The roadmap artifact
+
+The roadmap is an ordinary Markdown file you author and keep under version control —
+there is no special tooling behind it. Put it where the sub-specs can find it:
+
+- For a feature-scoped epic: `specs/<epic-slug>/roadmap.md`.
+- For a larger, cross-cutting epic: a top-level `ROADMAP.md`.
+
+Each roadmap entry carries a stable id (used later for linking), a name, its intent,
+its scope boundary, its dependencies, a status, and — once the sub-spec exists — a
+link to it. A minimal template:
+
+```markdown
+# Roadmap: <epic name>
+
+<One or two sentences: what the epic is and why it is being decomposed.>
+
+**Status legend**: planned · in-progress · done
+
+| ID | Sub-feature | Intent | Scope boundary | Depends on | Status | Sub-spec |
+|----|-------------|--------|----------------|-----------|--------|----------|
+| R1 | <name>      | <one line> | <in / deferred> | —      | planned | — |
+| R2 | <name>      | <one line> | <in / deferred> | R1     | planned | — |
+| R3 | <name>      | <one line> | <in / deferred> | R1     | planned | — |
+```
+
+Keep the `ID` column immutable once a sub-spec references it — it is the anchor for
+traceability. Fill in the `Sub-spec` column with the path to each sub-feature's spec
+directory as you create it, and update `Status` as work progresses.
+
+## Specifying each sub-feature
+
+With the roadmap in hand, work through the entries one at a time using the normal
+Spec Kit flow — nothing new to learn:
+
+1. Pick the next roadmap entry whose dependencies are already `done` (or have none).
+2. Run `/speckit.specify` for just that slice, describing only its intent and scope
+   from the roadmap entry. Because the slice is bounded, its spec, plan, and tasks
+   stay well within the context window.
+3. Run `/speckit.plan`, `/speckit.tasks`, and `/speckit.implement` for that slice as
+   usual.
+4. Mark the roadmap entry `done` and move to the next one.
+
+Each slice is a complete, independent Spec Kit feature with its own
+`spec.md`/`plan.md`/`tasks.md`. The roadmap is what ties them together.
+
+## Linking sub-specs to the roadmap
+
+To keep scope and intent from drifting across separate runs, every sub-spec
+references its roadmap entry, and the roadmap links back — a simple, greppable,
+bidirectional convention:
+
+- **Sub-spec → roadmap.** In the sub-feature's `spec.md`, name the parent roadmap
+  and entry id in the `Input` / summary line, for example:
+
+  ```markdown
+  **Input**: Parent roadmap: `specs/<epic>/roadmap.md` → entry **R3**. <feature description>
+  ```
+
+- **Roadmap → sub-spec.** In the roadmap table, set the entry's `Sub-spec` column to
+  the sub-feature's directory, e.g. `specs/<epic>-part-3/`.
+
+Because both directions are plain text, you can trace any sub-spec back to its place
+in the epic (and find its siblings) with a quick search — no tooling, no metadata
+schema.
+
+## Keeping the roadmap and sub-specs in sync
+
+The roadmap is a living document. As you learn more, keep it and the sub-specs
+aligned:
+
+- **Roadmap first, then reconcile.** When scope shifts, update the roadmap entry
+  first, then update any sub-specs it affects. The roadmap is the source of truth for
+  how the epic is divided.
+- **Respect dependencies and ordering.** If a slice depends on another, build the
+  prerequisite first and cross-reference the dependent sub-spec so the relationship
+  is visible from both sides.
+- **Recurse when a slice is still too big.** If a sub-feature turns out to be too
+  large to specify in one cycle, give it its own roadmap and decompose it further —
+  the same approach applies one level down. Recursion adds overhead, so only go as
+  deep as the context problem actually requires.
+
+## Worked example
+
+Suppose the epic is **"Add a self-service billing portal"** — far too large for a
+single cycle. The roadmap pass breaks it into three independently-specifiable
+slices.
+
+`specs/billing-portal/roadmap.md`:
+
+```markdown
+# Roadmap: Self-service billing portal
+
+Let customers view invoices, manage payment methods, and change plans without
+contacting support. Too large for one cycle, so it is split into independent slices.
+
+**Status legend**: planned · in-progress · done
+
+| ID | Sub-feature        | Intent                                   | Scope boundary                              | Depends on | Status  | Sub-spec |
+|----|--------------------|------------------------------------------|---------------------------------------------|-----------|---------|----------|
+| R1 | Invoice history    | Customers view and download past invoices | Read-only; no payment actions               | —         | done    | specs/billing-invoices/ |
+| R2 | Payment methods    | Add, remove, and set a default card       | No plan changes; assumes invoices exist     | R1        | in-progress | specs/billing-payment-methods/ |
+| R3 | Plan changes       | Upgrade/downgrade the subscription plan   | Uses R2's default payment method            | R1, R2    | planned | — |
+```
+
+Each slice is then specified on its own. For example, the **R2** sub-feature's
+`spec.md` opens with a back-reference:
+
+```markdown
+# Feature Specification: Billing — payment methods
+
+**Input**: Parent roadmap: `specs/billing-portal/roadmap.md` → entry **R2**.
+Let customers add, remove, and set a default payment method in the billing portal.
+```
+
+From here a reader can trace **R2** back to the roadmap, see that it depends on
+**R1** (invoice history, already `done`), and see that **R3** (plan changes) is
+waiting on it. Building R1, then R2, then R3 keeps every run small while the roadmap
+preserves the shape of the whole epic.
+
+## For automation (optional)
+
+If you would rather automate roadmap capture and consistency checks than maintain
+the file by hand, the community-maintained
+[Spec Roadmap extension](https://github.com/srobroek/speckit-roadmap) explores that
+direction. It is a third-party extension and is not required — the manual convention
+above is enough on its own.

docs/installation.md

@@ -77,9 +77,9 @@ specify init <project_name> --integration pi
 specify init <project_name> --integration omp
 ```
 
-### Specify Script Type (Shell vs PowerShell)
+### Specify Script Type (Shell, PowerShell, or Python)
 
-All automation scripts now have both Bash (`.sh`) and PowerShell (`.ps1`) variants.
+Automation scripts are available as Bash (`.sh`), PowerShell (`.ps1`), and Python (`.py`) variants.
 
 Auto behavior:
 
@@ -92,6 +92,7 @@ Force a specific script type:
 ```bash
 specify init <project_name> --script sh
 specify init <project_name> --script ps
+specify init <project_name> --script py
 ```
 
 ### Ignore Agent Tools Check
@@ -131,6 +132,7 @@ Scripts are installed into a variant subdirectory matching the chosen script typ
 
 - `.specify/scripts/bash/` — contains `.sh` scripts (default on Linux/macOS)
 - `.specify/scripts/powershell/` — contains `.ps1` scripts (default on Windows)
+- `.specify/scripts/python/` — contains `.py` scripts (chosen with `--script py`; also installs the platform shell fallback)
 
 ## Troubleshooting
 

docs/local-development.md

@@ -120,10 +120,10 @@ generated metadata, then add the import and `_register()` call in
 
 ## 7. Run Lint / Basic Checks
 
-CI enforces `ruff check src/` (see `.github/workflows/test.yml`), so run it locally before pushing:
+CI enforces `ruff check src tests` (see `.github/workflows/test.yml`), so run it locally before pushing:
 
 ```bash
-uvx ruff check src/
+uvx ruff check src tests
 ```
 
 You can also quickly sanity check importability:

docs/reference/core.md

@@ -12,7 +12,7 @@ specify init [<project_name>]
 | ------------------------ | ------------------------------------------------------------------------ |
 | `--integration <key>`    | AI coding agent integration to use (e.g. `copilot`, `claude`, `gemini`). See the [Integrations reference](integrations.md) for all available keys |
 | `--integration-options`  | Options for the integration (e.g. `--integration-options="--commands-dir .myagent/cmds"`) |
-| `--script sh\|ps`        | Script type: `sh` (bash/zsh) or `ps` (PowerShell)                       |
+| `--script sh\|ps\|py`    | Script type: `sh` (bash/zsh), `ps` (PowerShell), or `py` (Python)       |
 | `--here`                 | Initialize in the current directory instead of creating a new one        |
 | `--force`                | Force merge/overwrite when initializing in an existing directory         |
 | `--ignore-agent-tools`   | Skip checks for AI coding agent CLI tools                                |

docs/reference/extensions.md

@@ -221,12 +221,14 @@ Each hook entry supports the following fields:
 | `command` | Extension command associated with the hook. |
 | `enabled` | Whether the hook is active. Hooks with `enabled: false` are skipped. |
 | `optional` | Whether the hook is optional. If `true`, the hook is presented with its `prompt` and can be skipped; if `false`, the hook is emitted as an automatic hook (includes `EXECUTE_COMMAND` markers). |
-| `priority` | Priority metadata for the hook. Values must be integers >= 1; invalid values fall back to the default priority `10`. Current command templates surface hooks in their configured YAML order and do not sort them by `priority`. |
+| `priority` | Priority metadata for the hook. Registered hook entries use integer values >= 1; entries installed from manifests default to `10` when no priority is declared. Current command templates surface hooks in their configured YAML order and do not sort them by `priority`. |
 | `prompt` | Message shown when asking whether to run an optional hook. |
 | `description` | Human-readable explanation of what the hook does. |
 | `condition` | Optional expression evaluated by `HookExecutor` (using `config.<path>` or `env.<VAR>` with `is set`, `==`, or `!=`). Current command templates do not evaluate conditions and skip hooks with a non-empty condition. |
 Hook event names identify when a hook is invoked. They generally use `before_<command>` or `after_<command>`, such as `before_implement`, `after_implement`, `before_tasks`, and `after_tasks`.
 
+Extension manifests reject invalid hook priorities during installation. For existing `.specify/extensions.yml` entries, `HookExecutor.get_hooks_for_event()` sorts with `normalize_priority()`: missing values, booleans, non-numeric values rejected by `int()`, and values less than `1` fall back to `10`; numeric strings and finite floats are coerced with `int()`, while non-finite floats are unsupported and may fail instead of falling back.
+
 `HookExecutor.get_hooks_for_event()` returns hooks ordered by `priority`, with lower values first. However, current command templates read hook lists directly and surface them in their configured YAML order rather than using priority ordering.
 
 ## FAQ

docs/reference/integrations.md

@@ -15,6 +15,7 @@ The Specify CLI supports a wide range of AI coding agents. When you run `specify
 | [Codex CLI](https://github.com/openai/codex)                                         | `codex`          | Skills-based integration; installs skills into `.agents/skills` and invokes them as `$speckit-<command>` |
 | [Cursor](https://cursor.sh/)                                                         | `cursor-agent`   |                                                                                                                                           |
 | [Devin for Terminal](https://cli.devin.ai/docs)                                      | `devin`          | Skills-based integration; installs skills into `.devin/skills/` and invokes them as `/speckit-<command>` |
+| [Factory Droid](https://docs.factory.ai/cli/getting-started/overview)               | `droid`          | Skills-based integration; installs skills into `.factory/skills/` and invokes them as `/speckit-<command>`                               |
 | [Firebender](https://firebender.com/)                                                | `firebender`     | IDE-based agent for Android Studio / IntelliJ                                                                                             |
 | [Forge](https://forgecode.dev/)                                                      | `forge`          |                                                                                                                                           |
 | [Gemini CLI](https://github.com/google-gemini/gemini-cli)                            | `gemini`         |                                                                                                                                           |
@@ -22,7 +23,7 @@ The Specify CLI supports a wide range of AI coding agents. When you run `specify
 | [Goose](https://goose-docs.ai/)                                                      | `goose`          | Uses YAML recipe format in `.goose/recipes/`                                                                                              |
 | [Grok Build](https://docs.x.ai/build/overview)                                       | `grok`           | Skills-based integration; installs skills into `.grok/skills` and invokes them as `/speckit-<command>`                                    |
 | [Hermes](https://github.com/NousResearch/hermes-agent)                               | `hermes`         | Skills-based integration; installs skills globally into `~/.hermes/skills/`                                                                |
-| [IBM Bob](https://www.ibm.com/products/bob)                                          | `bob`            | IDE-based agent                                                                                                                           |
+| [IBM Bob](https://www.ibm.com/products/bob)                                          | `bob`            | Skills-based integration by default; installs skills as `speckit-<command>/SKILL.md` under `.bob/skills/` and invokes them as `/speckit-<command>`. Pass `--integration-options="--legacy-commands"` to scaffold the deprecated Bob 1.x layout (`.bob/commands/*.md`) instead; that flag will be removed in a future release. Existing legacy installs can migrate with `specify integration upgrade bob --integration-options="--skills"`, which converts them to the skills layout and removes the old command files. If preset overrides are installed, the migration is rejected with an actionable error (preset artifacts cannot yet be reconciled across a layout change) — remove the preset(s), migrate, then reinstall them. |
 | [Junie](https://junie.jetbrains.com/)                                                | `junie`          |                                                                                                                                           |
 | [Kilo Code](https://github.com/Kilo-Org/kilocode)                                    | `kilocode`       |                                                                                                                                           |
 | [Kimi Code](https://code.kimi.com/)                                                  | `kimi`           | Skills-based integration; installs into `.kimi-code/skills/`. `--migrate-legacy` moves old `.kimi/skills/` installs to the new paths |
@@ -48,7 +49,11 @@ The Specify CLI supports a wide range of AI coding agents. When you run `specify
 specify integration list
 ```
 
-Shows all available integrations, which one is currently installed, and whether each requires a CLI tool or is IDE-based.
+| Option      | Description                                                                                                             |
+| ----------- | ----------------------------------------------------------------------------------------------------------------------- |
+| `--catalog` | Also browse the catalog (built-in **and** community). Community integrations that are not built in are only shown here.  |
+
+Shows the built-in integrations, which one is currently installed, and whether each requires a CLI tool or is IDE-based.
 When multiple integrations are installed, the list marks the default integration separately from the other installed integrations.
 The list also shows whether each built-in integration is declared multi-install safe.
 

docs/reference/workflows.md

@@ -91,8 +91,192 @@ specify workflow add <source>
 | `--dev`         | Install from a local workflow YAML file or directory   |
 | `--from <url>`  | Install from a custom URL (`<source>` names the expected workflow ID) |
 
-Installs a workflow from the catalog, a URL (HTTPS required), or a local file path.
+Installs a workflow from the catalog, a URL (HTTPS required), a local YAML file, or a local directory containing `workflow.yml`.
 
+## Workflow Overlays
+
+Workflow overlays let a project extend or override an installed workflow without editing the installed `workflow.yml`.  This keeps local customizations safe across `specify bundle update` or `specify workflow add` upgrades.
+
+When `specify workflow run <workflow-id>` loads a workflow, the engine composes the base workflow with all enabled overlays for that workflow id.  The result is validated like any other workflow definition.
+
+### How Overlays Work
+
+An overlay is a YAML file that declares a set of edit operations against the step list of a base workflow. Overlays use lower-wins precedence: higher priority numbers are applied first and lower numbers last. Equal-priority overlays are applied alphabetically by ID, with the last ID winning conflicts.
+
+Project overlay files live at:
+
+| Location | Purpose |
+| --- | --- |
+| `.specify/workflows/overlays/<id>/*.yml` | Project-local customizations |
+
+### Overlay File Format
+
+The recommended edit format uses the operation name as the key and the anchor step id as the value:
+
+```yaml
+id: "my-overlay"
+extends: "speckit"
+priority: 10
+enabled: true
+edits:
+  - insert_after: implement
+    step:
+      id: run-lint
+      type: shell
+      run: "ruff check src/"
+
+  - replace: review-spec
+    step:
+      id: review-spec
+      type: gate
+      message: "Review the generated spec (overlay override)."
+      options: [approve, reject]
+      on_reject: abort
+```
+
+The explicit form is also supported:
+
+```yaml
+edits:
+  - operation: insert_after
+    anchor: implement
+    step:
+      id: run-lint
+      type: shell
+      run: "ruff check src/"
+```
+
+#### Fields
+
+| Field | Required | Description |
+| --- | --- | --- |
+| `id` | yes | Identifier for this overlay. Used in `specify workflow overlay *` commands. Must be lowercase letters, digits, and hyphens only; no dots, underscores, path separators, or `overlays`. |
+| `extends` | yes | The workflow id this overlay applies to. Uses the same safe-id format as `id`; `overlays`, `runs`, and `steps` are reserved. |
+| `priority` | no | Integer; defaults to `10`. Lower values have higher precedence and win conflicts. Missing or invalid values fall back to `10`. |
+| `enabled` | no | Boolean. Defaults to `true`. Disabled overlays are ignored. |
+| `edits` | yes | Non-empty list of edit operations. |
+
+#### Edit Operations
+
+| Operation | `step` required | Effect |
+| --- | --- | --- |
+| `insert_after` | yes | Insert `step` immediately after the anchor step. |
+| `insert_before` | yes | Insert `step` immediately before the anchor step. |
+| `replace` | yes | Replace the anchor step with `step`. |
+| `remove` | no | Remove the anchor step from the list. |
+
+The `anchor` is the `id` of a step in the base workflow.  Anchors are resolved recursively inside `then`, `else`, `steps`, `cases.*`, and `default` blocks, so nested base steps can also be targeted.  Fan-out templates (`step` inside a `fan-out` step) are **not** valid anchors.
+
+Step ids must not contain `:` — that character is reserved for engine-generated nested ids.
+
+### Overlay CLI Commands
+
+#### Add a Project Overlay
+
+```bash
+specify workflow overlay add <path-to-overlay.yml> --priority <n>
+```
+
+Validates the overlay file and copies it to `.specify/workflows/overlays/<extends>/<id>.yml`. `--priority` defaults to `10` and overrides the `priority` field in the file.
+
+#### List Overlays
+
+```bash
+specify workflow overlay list <workflow-id>
+```
+
+Shows all overlays for the workflow, ordered by resolver precedence. Disabled overlays are marked as disabled in the listing and are ignored during workflow resolution.
+
+#### Change Priority
+
+```bash
+specify workflow overlay set-priority <workflow-id> <overlay-id> <n>
+```
+
+#### Enable or Disable
+
+```bash
+specify workflow overlay disable <workflow-id> <overlay-id>
+specify workflow overlay enable <workflow-id> <overlay-id>
+```
+
+#### Remove
+
+```bash
+specify workflow overlay remove <workflow-id> <overlay-id>
+```
+
+Removes the project overlay file.
+
+#### Inspect the Composed Workflow
+
+```bash
+specify workflow resolve <workflow-id>
+```
+
+Prints the layer stack (base + overlays) and the source attribution for each step after composition.  Useful for debugging which overlay contributed or overrode a step.
+
+### Example: Adding Automated Linting after Implementation
+
+Given the built-in `speckit` workflow, create `project-overlay.yml`:
+
+```yaml
+id: "add-lint"
+extends: "speckit"
+priority: 10
+edits:
+  - insert_after: implement
+    step:
+      id: run-lint
+      type: shell
+      run: "ruff check src/"
+```
+
+Install it:
+
+```bash
+specify workflow overlay add project-overlay.yml --priority 10
+```
+
+Run the workflow:
+
+```bash
+specify workflow run speckit -i spec="Build a kanban board"
+```
+
+The composed workflow will now run the full SDD cycle and execute `ruff check src/` automatically after the `implement` step.
+
+### Example: Replacing a Gate
+
+```yaml
+id: "skip-plan-review"
+extends: "speckit"
+priority: 5
+edits:
+  - replace: review-plan
+    step:
+      id: review-plan
+      type: command
+      command: speckit.plan
+      input:
+        args: "{{ inputs.spec }}"
+```
+
+Lower priority values have higher precedence. Change this overlay to `priority: 5` if it must win a conflict with the `add-lint` overlay above. It replaces the `review-plan` gate with a non-interactive command.
+
+### Interaction with Bundles and Updates
+
+`specify workflow add <local-directory>` installs `workflow.yml` from the local directory into `.specify/workflows/<id>/`.
+
+When an installed workflow is refreshed or reinstalled, project overlays in `.specify/workflows/overlays/<id>/` are preserved because they live outside the installed workflow directory.
+
+### Limitations
+
+- Overlays operate on the step list only.  They cannot change workflow metadata (name, description, inputs, `requires`) or expression logic.
+- Fan-out templates cannot be used as anchors.
+- An overlay that targets a step id that does not exist in the base workflow will raise a validation error when the workflow is resolved.
+- Overlays cannot target steps added by other overlays.
+- Overlays cannot add new inputs or change the input schema of the base workflow.
 ## Update Workflows
 
 ```bash

docs/toc.yml

@@ -55,6 +55,8 @@
       href: concepts/spec-persistence.md
     - name: Handling Complex Features
       href: concepts/complex-features.md
+    - name: Spec of Specs
+      href: concepts/spec-of-specs.md
 
 # Development workflows
 - name: Development

docs/upgrade.md

@@ -12,7 +12,7 @@
 | **CLI Tool — pin a version** | `specify self upgrade --tag vX.Y.Z[suffix]` | Upgrade to a specific release tag instead of the latest stable. Suffixes are limited to dev, alpha/beta/rc, and/or build metadata forms. |
 | **CLI Tool — manual fallback** | `uv tool install specify-cli --force --from git+https://github.com/github/spec-kit.git@vX.Y.Z` | When `specify self upgrade` isn't available (older installs) or when you want explicit control. |
 | **CLI Tool — manual fallback (pipx)** | `pipx install --force git+https://github.com/github/spec-kit.git@vX.Y.Z` | Same as above, for pipx installs. |
-| **Project Files** | `specify init --here --force --integration <your-agent>` | Update slash commands, templates, and scripts in your project |
+| **Project Files** | Run `specify integration upgrade <key>`, then `specify extension update` | Refresh installed integration files and extensions in your project |
 | **Both** | Run CLI upgrade, then project update | Recommended for major version updates |
 
 ---
@@ -89,91 +89,94 @@ specify self check
 
 ## Part 2: Updating Project Files
 
-When Spec Kit releases new features (like new slash commands or updated templates), you need to refresh your project's Spec Kit files.
+When Spec Kit releases new features (like new slash commands, updated templates, or extension changes), you need to refresh the Spec Kit files that were installed into your project.
 
 ### What gets updated?
 
-Running `specify init --here --force` will update:
+For existing Spec Kit projects, use the manifest-aware upgrade path first:
 
-- ✅ **Slash command files** (`.claude/commands/`, `.github/prompts/`, etc.)
-- ✅ **Script files** (`.specify/scripts/`) — **only with `--force`**; without it, only missing files are added
-- ✅ **Template files** (`.specify/templates/`) — **only with `--force`**; without it, only missing files are added
-- ✅ **Shared memory files** (`.specify/memory/`) - **⚠️ See warnings below**
+- ✅ **Integration command/skill files** (`.claude/skills/`, `.github/prompts/`, `.agents/skills/`, etc.)
+- ✅ **Managed shared scripts and templates** (`.specify/scripts/`, `.specify/templates/`) when they are unchanged from the previous managed copy
+- ✅ **Installed extensions** when you run `specify extension update`
+
+The integration upgrade command uses the install manifest to detect local edits. If a managed integration file was modified after install, the command stops and asks you to inspect the change or rerun with `--force`.
 
 ### What stays safe?
 
-These files are **never touched** by the upgrade—the template packages don't even contain them:
+These files are **never touched** by the manifest-aware integration/extension upgrade path:
 
 - ✅ **Your specifications** (`specs/001-my-feature/spec.md`, etc.) - **CONFIRMED SAFE**
 - ✅ **Your implementation plans** (`specs/001-my-feature/plan.md`, `tasks.md`, etc.) - **CONFIRMED SAFE**
+- ✅ **Your constitution** (`.specify/memory/constitution.md`) when using `specify integration upgrade`
 - ✅ **Your source code** - **CONFIRMED SAFE**
 - ✅ **Your git history** - **CONFIRMED SAFE**
 
 The `specs/` directory is completely excluded from template packages and will never be modified during upgrades.
 
-### Update command
+### 1. Check installed integrations
 
 Run this inside your project directory:
 
+```bash
+specify integration status
+```
+
+This reports the default integration, all installed integrations, and any modified or missing managed files. You can also inspect `.specify/integration.json`; installed integrations are listed under `installed_integrations`.
+
+### 2. Upgrade each installed integration
+
+Run this inside your project directory:
+
+```bash
+specify integration upgrade <key>
+```
+
+Replace `<key>` with an installed integration key such as `copilot`, `claude`, or `codex`. In projects with multiple installed integrations, run the command once per installed key.
+
+**Example:**
+
+```bash
+specify integration upgrade claude
+specify integration upgrade codex
+```
+
+See the [integration reference](reference/integrations.md#upgrade-an-integration) for options such as `--script`, `--integration-options`, and `--force`.
+
+### 3. Update installed extensions
+
+Run:
+
+```bash
+specify extension update
+```
+
+With no extension argument, this updates all installed extensions. Use `specify extension update <extension-id-or-name>` to update only one extension. See the [extensions reference](reference/extensions.md#update-extensions) for details.
+
+### Fallback: re-run init
+
+If a project predates manifests, has missing integration metadata, or needs a broader recovery, you can still re-run init:
+
 ```bash
 specify init --here --force --integration <your-agent>
 ```
 
-Replace `<your-agent>` with your AI coding agent. Refer to this list of [Supported AI Coding Agent Integrations](reference/integrations.md)
-
-**Example:**
-
-```bash
-specify init --here --force --integration copilot
-```
-
-### Understanding the `--force` flag
-
-Without `--force`, the CLI warns you and asks for confirmation:
-
-```text
-Warning: Current directory is not empty (25 items)
-Template files will be merged with existing content and may overwrite existing files
-Proceed? [y/N]
-```
-
-With `--force`, it skips the confirmation and proceeds immediately. It also **overwrites shared infrastructure files** (`.specify/scripts/` and `.specify/templates/`) with the latest versions from the installed Spec Kit release.
-
-Without `--force`, shared infrastructure files that already exist are skipped — the CLI will print a warning listing the skipped files so you know which ones were not updated.
-
-**Important: Your `specs/` directory is always safe.** The `--force` flag only affects template files (commands, scripts, templates, memory). Your feature specifications, plans, and tasks in `specs/` are never included in upgrade packages and cannot be overwritten.
-
----
+Use this as an escape hatch rather than the default project-file upgrade path. It refreshes the selected integration and shared project scaffolding, but it does not use the same per-integration manifest checks before overwriting files.
 
 ## ⚠️ Important Warnings
 
-### 1. Constitution file will be overwritten
+### 1. Constitution file and memory customizations
 
-**Known issue:** `specify init --here --force` currently overwrites `.specify/memory/constitution.md` with the default template, erasing any customizations you made.
+`specify integration upgrade <key>` does not update `.specify/memory/constitution.md`.
 
-**Workaround:**
+The fallback `specify init --here --force --integration <your-agent>` path also preserves an existing `.specify/memory/constitution.md`; if the file is missing, init creates it from the current constitution template. You do not need a constitution backup/restore step for the manifest-aware upgrade path.
 
-```bash
-# 1. Back up your constitution before upgrading
-cp .specify/memory/constitution.md .specify/memory/constitution-backup.md
+As with any broad fallback refresh, commit or back up local customizations before using `init --here --force` so you can review the resulting diff.
 
-# 2. Run the upgrade
-specify init --here --force --integration copilot
+### 2. Custom integration, script, or template modifications
 
-# 3. Restore your customized constitution
-mv .specify/memory/constitution-backup.md .specify/memory/constitution.md
-```
+`specify integration upgrade <key>` blocks when manifest-tracked integration files were modified locally, unless you pass `--force`.
 
-Or use git to restore it:
-
-```bash
-# After upgrade, restore from git history
-git restore .specify/memory/constitution.md
-```
-
-### 2. Custom script or template modifications
-
-If you customized files in `.specify/scripts/` or `.specify/templates/`, the `--force` flag will overwrite them. Back them up first:
+Shared scripts and templates are refreshed when they still match the previously recorded managed copy. Local customizations are preserved unless you explicitly use a force/refresh option that overwrites them. If you customized files in `.specify/scripts/` or `.specify/templates/`, commit or back them up first:
 
 ```bash
 # Back up custom templates and scripts
@@ -215,29 +218,29 @@ Restart your IDE to refresh the command list.
 # Upgrade CLI (auto-detects uv tool vs pipx install)
 specify self upgrade
 
-# Update project files to get new commands
-specify init --here --force --integration copilot
+# Inspect installed integrations
+specify integration status
 
-# Restore your constitution if customized
-git restore .specify/memory/constitution.md
+# Update project files to get new commands
+specify integration upgrade <key>
+specify extension update
 ```
 
 ### Scenario 2: "I customized templates and constitution"
 
 ```bash
-# 1. Back up customizations
-cp .specify/memory/constitution.md /tmp/constitution-backup.md
+# 1. Commit or back up customizations
+git status
 cp -r .specify/templates /tmp/templates-backup
 
 # 2. Upgrade CLI
 specify self upgrade
 
-# 3. Update project
-specify init --here --force --integration copilot
+# 3. Use the manifest-aware project update first
+specify integration upgrade <key>
+specify extension update
 
-# 4. Restore customizations
-mv /tmp/constitution-backup.md .specify/memory/constitution.md
-# Manually merge template changes if needed
+# 4. If the upgrade reports modified managed files, inspect the diff before using --force
 ```
 
 ### Scenario 3: "I see duplicate slash commands in my IDE"
@@ -262,14 +265,14 @@ rm speckit.old-command-name.md
 The git extension is now opt-in, so upgrades do not install it unless you add it explicitly.
 
 ```bash
-# Manually back up files you customized
-cp .specify/memory/constitution.md .specify/memory/constitution.backup.md
+# Upgrade CLI
+specify self upgrade
 
-# Run upgrade
-specify init --here --force --integration copilot
+# Refresh integration files and installed extensions
+specify integration upgrade <key>
+specify extension update
 
-# Restore customizations
-mv .specify/memory/constitution.backup.md .specify/memory/constitution.md
+# The git extension is not added unless you run `specify extension add git`
 ```
 
 If you later decide you want the git extension's commands and hooks, install it explicitly:
@@ -315,19 +318,21 @@ Alternatively, run the `/speckit.specify` command which creates `.specify/featur
    - Codex requires `CODEX_HOME` environment variable
    - Some agents need workspace restart or cache clearing
 
-### "I lost my constitution customizations"
+### "Will init overwrite my constitution customizations?"
 
-**Fix:** Restore from git or backup:
+Current `specify init --here --force` preserves an existing `.specify/memory/constitution.md`; it creates the file from the template only when it is missing.
+
+If you previously lost constitution changes through an older workflow or manual replacement, restore from git or backup:
 
 ```bash
-# If you committed before upgrading
+# If you committed the customized constitution
 git restore .specify/memory/constitution.md
 
 # If you backed up manually
 cp /tmp/constitution-backup.md .specify/memory/constitution.md
 ```
 
-**Prevention:** Always commit or back up `constitution.md` before upgrading.
+**Prevention:** Use `specify integration upgrade <key>` for routine project-file updates. If you need the fallback `specify init --here --force` path, commit first so you can review the full diff afterward.
 
 ### "Warning: Current directory is not empty"
 
@@ -354,7 +359,7 @@ Only Spec Kit infrastructure files:
 - Agent command files (`.claude/commands/`, `.github/prompts/`, etc.)
 - Scripts in `.specify/scripts/`
 - Templates in `.specify/templates/`
-- Memory files in `.specify/memory/` (including constitution)
+- Missing memory files such as `.specify/memory/constitution.md` may be created from templates; an existing constitution is preserved
 
 **What stays untouched:**
 
@@ -365,7 +370,7 @@ Only Spec Kit infrastructure files:
 
 **How to respond:**
 
-- **Type `y` and press Enter** - Proceed with the merge (recommended if upgrading)
+- **Type `y` and press Enter** - Proceed with the merge when using the fallback init path
 - **Type `n` and press Enter** - Cancel the operation
 - **Use `--force` flag** - Skip this confirmation entirely:
 
@@ -375,11 +380,11 @@ Only Spec Kit infrastructure files:
 
 **When you see this warning:**
 
-- ✅ **Expected** when upgrading an existing Spec Kit project
+- ✅ **Expected** when using the fallback init path in an existing Spec Kit project
 - ✅ **Expected** when adding Spec Kit to an existing codebase
 - ⚠️ **Unexpected** if you thought you were creating a new project in an empty directory
 
-**Prevention tip:** Before upgrading, commit or back up your `.specify/memory/constitution.md` if you customized it.
+**Prevention tip:** Before using the fallback init path, commit your current work so any refreshed files are easy to review or restore.
 
 ### "CLI upgrade doesn't seem to work"
 
@@ -418,14 +423,15 @@ uv tool install specify-cli --from git+https://github.com/github/spec-kit.git
 
 ### "Do I need to run specify every time I open my project?"
 
-**Short answer:** No, you only run `specify init` once per project (or when upgrading).
+**Short answer:** No, you only run `specify init` once per project, or later as a fallback recovery path.
 
 **Explanation:**
 
 The `specify` CLI tool is used for:
 
 - **Initial setup:** `specify init` to bootstrap Spec Kit in your project
-- **Upgrades:** `specify init --here --force` to update templates and commands
+- **Routine project-file upgrades:** `specify integration upgrade <key>` and `specify extension update`
+- **Fallback recovery:** `specify init --here --force` when integration metadata is missing or the manifest-aware path cannot be used
 - **Diagnostics:** `specify check` to verify tool installation
 
 Once you've run `specify init`, the slash commands (like `/speckit.specify`, `/speckit.plan`, etc.) are **permanently installed** in your project's agent folder (`.claude/`, `.github/prompts/`, `.pi/prompts/`, `.omp/commands/`, etc.). Your AI coding agent reads these command files directly—no need to run `specify` again.

extensions/EXTENSION-DEVELOPMENT-GUIDE.md

@@ -252,6 +252,7 @@ Use standard Markdown with special placeholders:
 
 - `$ARGUMENTS`: User-provided arguments
 - `{SCRIPT}`: Replaced with script path during registration
+- `__SPECKIT_COMMAND_<NAME>__`: Replaced with the invocation of another command, rendered using the active integration's separator (see [Referencing other commands](#referencing-other-commands))
 
 **Example**:
 
@@ -267,6 +268,40 @@ echo "Running with args: $args"
 ```
 ````
 
+### Referencing other commands
+
+A command body is a *template* that Spec Kit renders once per agent. Different agents invoke commands with different surface syntax — for example `/speckit.plan` (dot separator) or `/speckit-plan` (hyphen separator). Some agents also use different prefixes in skills mode (e.g. Kimi `/skill:speckit-plan`, Codex/ZCode `$speckit-plan`). So when you reference a sibling command from a body, **do not hard-code a literal invocation** like `/speckit.my-ext.prepare`. A literal is correct for exactly one agent and breaks on the rest.
+
+Instead use the agent-neutral token `__SPECKIT_COMMAND_<NAME>__`. Spec Kit resolves it to a `/speckit<separator>...` invocation using the active integration's `invoke_separator` (and integrations may post-process that further in skills output).
+
+Encode the command name in upper case, dropping the `speckit.` prefix and turning each dotted segment separator into an underscore:
+
+| Command file | Token |
+| --- | --- |
+| `speckit.plan.md` | `__SPECKIT_COMMAND_PLAN__` |
+| `speckit.bug.fix.md` | `__SPECKIT_COMMAND_BUG_FIX__` |
+| `speckit.git.commit.md` | `__SPECKIT_COMMAND_GIT_COMMIT__` |
+
+The resolver maps each underscore back to the active agent's separator, so use tokens to reference commands whose name segments are single words. (Command names are dotted segments like `git.commit`; the token scheme rebuilds those dots and does not carry hyphens within a segment.)
+
+**Example** — a command body that points the user at the next step:
+
+```markdown
+Once the assessment exists, the next step is `__SPECKIT_COMMAND_BUG_FIX__ slug=<slug>`.
+```
+
+This renders as `/speckit.bug.fix slug=<slug>` for a slash-based agent, `/speckit-bug-fix slug=<slug>` for a skills-based agent, and so on — the author writes it once and it stays portable. The first-party `bug` and `git` extensions use this token exclusively; see `extensions/bug/commands/` for working examples.
+
+> **Current limitation — skills mode.** Token resolution runs in the
+> command-rendering path (`CommandRegistrar`), so it applies when an extension
+> installs *command files*. It does **not** yet run when an extension is
+> registered as *skills* for a skills-based agent: `_register_extension_skills`
+> resolves placeholders and post-processes content but never calls
+> `resolve_command_refs`, so a `__SPECKIT_COMMAND_<NAME>__` token reaches
+> agents such as Codex, ZCode, and Kimi verbatim in that mode. Until that
+> rendering step lands, prefer the token for command-file extensions and avoid
+> relying on it inside skill bodies destined for skills-based agents.
+
 ### Script Path Rewriting
 
 Extension commands use relative paths that get rewritten during registration:

extensions/catalog.community.json

@@ -1,6 +1,6 @@
 {
   "schema_version": "1.0",
-  "updated_at": "2026-07-17T00:00:00Z",
+  "updated_at": "2026-07-21T00:00:00Z",
   "catalog_url": "https://raw.githubusercontent.com/github/spec-kit/main/extensions/catalog.community.json",
   "extensions": {
     "aide": {
@@ -2029,6 +2029,40 @@
       "created_at": "2026-06-01T00:00:00Z",
       "updated_at": "2026-06-22T00:00:00Z"
     },
+    "linear-weave": {
+      "name": "Linear Weave",
+      "id": "linear-weave",
+      "description": "Weave Spec Kit into Linear: pull requirements, mirror tasks.md into sub-issues, sync statuses.",
+      "author": "Tony Woodhouse",
+      "version": "1.0.0",
+      "download_url": "https://github.com/tonydwoodhouse/spec-kit-linear-weave/archive/refs/tags/v1.0.0.zip",
+      "repository": "https://github.com/tonydwoodhouse/spec-kit-linear-weave",
+      "homepage": "https://github.com/tonydwoodhouse/spec-kit-linear-weave",
+      "documentation": "https://github.com/tonydwoodhouse/spec-kit-linear-weave#readme",
+      "changelog": "https://github.com/tonydwoodhouse/spec-kit-linear-weave/blob/main/CHANGELOG.md",
+      "license": "MIT",
+      "category": "integration",
+      "effect": "read-write",
+      "requires": {
+        "speckit_version": ">=0.13.0,<1.0.0",
+        "tools": [{ "name": "linear-mcp", "required": true }]
+      },
+      "provides": {
+        "commands": 5,
+        "hooks": 5
+      },
+      "tags": [
+        "linear",
+        "issue-tracking",
+        "integration",
+        "workflow"
+      ],
+      "verified": false,
+      "downloads": 0,
+      "stars": 0,
+      "created_at": "2026-07-21T00:00:00Z",
+      "updated_at": "2026-07-21T00:00:00Z"
+    },
     "loop": {
       "name": "Loop Engineering",
       "id": "loop",
@@ -2683,10 +2717,10 @@
     "okf": {
       "name": "OKF Knowledge Bundle Generator",
       "id": "okf",
-      "description": "Generates and maintains an Open Knowledge Format (OKF v0.1) knowledge bundle from a source-code repository.",
+      "description": "Generates and maintains an Open Knowledge Format (OKF v0.1) knowledge bundle from a source-code repository, mining git history for significance and rationale, and resolving open questions with the user.",
       "author": "Alex Punnen",
-      "version": "0.2.0",
-      "download_url": "https://github.com/alexcpn/speckit_ofk/archive/refs/tags/v0.2.0.zip",
+      "version": "0.3.0",
+      "download_url": "https://github.com/alexcpn/speckit_ofk/archive/refs/tags/v0.3.0.zip",
       "repository": "https://github.com/alexcpn/speckit_ofk",
       "homepage": "https://github.com/alexcpn/speckit_ofk",
       "documentation": "https://github.com/alexcpn/speckit_ofk/blob/main/README.md",
@@ -2698,7 +2732,7 @@
         "speckit_version": ">=0.12.0"
       },
       "provides": {
-        "commands": 3,
+        "commands": 4,
         "hooks": 0
       },
       "tags": [
@@ -2712,7 +2746,7 @@
       "downloads": 0,
       "stars": 0,
       "created_at": "2026-07-17T00:00:00Z",
-      "updated_at": "2026-07-17T00:00:00Z"
+      "updated_at": "2026-07-21T00:00:00Z"
     },
     "onboard": {
       "name": "Onboard",
@@ -4342,6 +4376,40 @@
       "created_at": "2026-05-20T00:00:00Z",
       "updated_at": "2026-05-20T00:00:00Z"
     },
+    "test-coverage-drift-control": {
+      "name": "Test Coverage Drift Control",
+      "id": "test-coverage-drift-control",
+      "description": "Generate incremental coverage drift reports and planned remediation tasks after implementation",
+      "author": "Igor Benicio de Mesquita",
+      "version": "0.3.0",
+      "download_url": "https://github.com/benizzio/spec-kit-test-coverage-drift-control/archive/refs/tags/v0.3.0.zip",
+      "repository": "https://github.com/benizzio/spec-kit-test-coverage-drift-control",
+      "homepage": "https://github.com/benizzio/spec-kit-test-coverage-drift-control",
+      "documentation": "https://github.com/benizzio/spec-kit-test-coverage-drift-control#readme",
+      "changelog": "https://github.com/benizzio/spec-kit-test-coverage-drift-control/blob/main/CHANGELOG.md",
+      "license": "MIT",
+      "category": "code",
+      "effect": "read-write",
+      "requires": {
+        "speckit_version": ">=0.2.0"
+      },
+      "provides": {
+        "commands": 2,
+        "hooks": 1
+      },
+      "tags": [
+        "analysis",
+        "coverage",
+        "testing",
+        "quality",
+        "maintenance"
+      ],
+      "verified": false,
+      "downloads": 0,
+      "stars": 0,
+      "created_at": "2026-07-21T00:00:00Z",
+      "updated_at": "2026-07-21T00:00:00Z"
+    },
     "time-machine": {
       "name": "Time Machine",
       "id": "time-machine",

extensions/git/scripts/powershell/create-new-feature-branch.ps1

@@ -565,6 +565,12 @@ if (-not $DryRun) {
     $env:SPECIFY_FEATURE = $branchName
 }
 
+# Build the PowerShell-idiomatic persist hint, mirroring the core
+# create-new-feature.ps1 twin (and the bash/python twins of this script), which
+# all emit "# To persist in your shell: ...".
+$quotedBranchName = "'" + $branchName.Replace("'", "''") + "'"
+$featureAssignment = '$env:SPECIFY_FEATURE = ' + $quotedBranchName
+
 if ($Json) {
     $obj = [PSCustomObject]@{
         BRANCH_NAME = $branchName
@@ -581,6 +587,6 @@ if ($Json) {
     Write-Output "BRANCH_NAME: $branchName"
     Write-Output "FEATURE_NUM: $featureNum"
     if (-not $DryRun) {
-        Write-Output "SPECIFY_FEATURE environment variable set to: $branchName"
+        Write-Output "# To persist in your shell: $featureAssignment"
     }
 }

integrations/catalog.json

@@ -1,6 +1,6 @@
 {
   "schema_version": "1.0",
-  "updated_at": "2026-07-15T00:00:00Z",
+  "updated_at": "2026-07-17T00:00:00Z",
   "catalog_url": "https://raw.githubusercontent.com/github/spec-kit/main/integrations/catalog.json",
   "integrations": {
     "claude": {
@@ -48,6 +48,15 @@
       "repository": "https://github.com/github/spec-kit",
       "tags": ["ide"]
     },
+    "droid": {
+      "id": "droid",
+      "name": "Factory Droid",
+      "version": "1.0.0",
+      "description": "Factory Droid CLI skills-based integration",
+      "author": "spec-kit-core",
+      "repository": "https://github.com/github/spec-kit",
+      "tags": ["cli", "skills", "factory"]
+    },
     "amp": {
       "id": "amp",
       "name": "Amp",
@@ -177,11 +186,11 @@
     "bob": {
       "id": "bob",
       "name": "IBM Bob",
-      "version": "1.0.0",
-      "description": "IBM Bob IDE integration",
+      "version": "2.0.0",
+      "description": "IBM Bob 2.0 IDE skills-based integration",
       "author": "spec-kit-core",
       "repository": "https://github.com/github/spec-kit",
-      "tags": ["ide", "ibm"]
+      "tags": ["ide", "ibm", "skills"]
     },
     "trae": {
       "id": "trae",

presets/catalog.community.json

@@ -1,6 +1,6 @@
 {
   "schema_version": "1.0",
-  "updated_at": "2026-07-17T00:00:00Z",
+  "updated_at": "2026-07-22T00:00:00Z",
   "catalog_url": "https://raw.githubusercontent.com/github/spec-kit/main/presets/catalog.community.json",
   "presets": {
     "a11y-governance": {
@@ -69,7 +69,7 @@
       "name": "AIDE In-Place Migration",
       "id": "aide-in-place",
       "version": "1.0.0",
-      "description": "Adapts the AIDE workflow for in-place technology migrations (X → Y pattern). Overrides vision, roadmap, progress, and work item commands with migration-specific guidance.",
+      "description": "Adapts the AIDE workflow for in-place technology migrations (X \u2192 Y pattern). Overrides vision, roadmap, progress, and work item commands with migration-specific guidance.",
       "author": "mnriem",
       "repository": "https://github.com/mnriem/spec-kit-presets",
       "download_url": "https://github.com/mnriem/spec-kit-presets/releases/download/aide-in-place-v1.0.0/aide-in-place.zip",
@@ -134,13 +134,13 @@
     "autonomous-run-governance": {
       "name": "Autonomous Run Governance",
       "id": "autonomous-run-governance",
-      "version": "0.2.2",
-      "description": "Adds permission-bounded, evidence-first governance for autonomous Spec Kit delivery with validated status, stop, resume, exact-head proof, closeout, and learner guidance.",
+      "version": "0.3.2",
+      "description": "Adds permission-bounded, evidence-first governance for complete autonomous Spec Kit delivery, including validated status, stop, explicit resume, exact-head proof, post-merge closeout, retrospective learning, and an optional policy-driven intake-review gate before feature creation.",
       "author": "Thorsten Hindermann",
       "repository": "https://github.com/hindermath/spec-kit-preset-autonomous-run-governance",
-      "download_url": "https://github.com/hindermath/spec-kit-preset-autonomous-run-governance/archive/refs/tags/v0.2.2.zip",
+      "download_url": "https://github.com/hindermath/spec-kit-preset-autonomous-run-governance/archive/refs/tags/v0.3.2.zip",
       "homepage": "https://github.com/hindermath/spec-kit-preset-autonomous-run-governance",
-      "documentation": "https://github.com/hindermath/spec-kit-preset-autonomous-run-governance/blob/v0.2.2/README.md",
+      "documentation": "https://github.com/hindermath/spec-kit-preset-autonomous-run-governance/blob/v0.3.2/README.md",
       "license": "MIT",
       "requires": {
         "speckit_version": ">=0.8.3"
@@ -155,10 +155,11 @@
         "governance",
         "evidence",
         "permissions",
-        "resume"
+        "resume",
+        "intake-review"
       ],
       "created_at": "2026-07-13T00:00:00Z",
-      "updated_at": "2026-07-17T00:00:00Z"
+      "updated_at": "2026-07-21T00:00:00Z"
     },
     "canon-core": {
       "name": "Canon Core",
@@ -363,6 +364,64 @@
       "created_at": "2026-05-05T08:00:00Z",
       "updated_at": "2026-06-22T00:00:00Z"
     },
+    "intake-authoring-governance": {
+      "name": "Intake Authoring Governance",
+      "id": "intake-authoring-governance",
+      "version": "0.1.0",
+      "description": "Creates traceable Spec Kit intake files and receipts from ordered text sources while preserving clarification, update, and delivery-authority boundaries.",
+      "author": "Thorsten Hindermann",
+      "repository": "https://github.com/hindermath/spec-kit-preset-intake-authoring-governance",
+      "download_url": "https://github.com/hindermath/spec-kit-preset-intake-authoring-governance/archive/refs/tags/v0.1.0.zip",
+      "homepage": "https://github.com/hindermath/spec-kit-preset-intake-authoring-governance",
+      "documentation": "https://github.com/hindermath/spec-kit-preset-intake-authoring-governance/blob/v0.1.0/README.md",
+      "license": "MIT",
+      "requires": {
+        "speckit_version": ">=0.8.3"
+      },
+      "provides": {
+        "templates": 7,
+        "commands": 2,
+        "scripts": 2
+      },
+      "tags": [
+        "intake",
+        "authoring",
+        "governance",
+        "traceability",
+        "clarification"
+      ],
+      "created_at": "2026-07-22T00:00:00Z",
+      "updated_at": "2026-07-22T00:00:00Z"
+    },
+    "intake-review-governance": {
+      "name": "Intake Review Governance",
+      "id": "intake-review-governance",
+      "version": "0.1.0",
+      "description": "Adds hash-bound review, repair, and status gates for single, series, and campaign intake files before interactive, autonomous, or parallel Spec Kit execution.",
+      "author": "Thorsten Hindermann",
+      "repository": "https://github.com/hindermath/spec-kit-preset-intake-review-governance",
+      "download_url": "https://github.com/hindermath/spec-kit-preset-intake-review-governance/archive/refs/tags/v0.1.0.zip",
+      "homepage": "https://github.com/hindermath/spec-kit-preset-intake-review-governance",
+      "documentation": "https://github.com/hindermath/spec-kit-preset-intake-review-governance/blob/v0.1.0/README.md",
+      "license": "MIT",
+      "requires": {
+        "speckit_version": ">=0.8.3"
+      },
+      "provides": {
+        "templates": 8,
+        "commands": 3,
+        "scripts": 2
+      },
+      "tags": [
+        "intake",
+        "review",
+        "governance",
+        "quality-gate",
+        "autonomous"
+      ],
+      "created_at": "2026-07-21T00:00:00Z",
+      "updated_at": "2026-07-21T00:00:00Z"
+    },
     "isaqb-architecture-governance": {
       "name": "iSAQB Architecture Governance",
       "id": "isaqb-architecture-governance",
@@ -480,6 +539,36 @@
       "created_at": "2026-04-09T00:00:00Z",
       "updated_at": "2026-04-09T00:00:00Z"
     },
+    "parallel-autonomous-run-governance": {
+      "name": "Parallel Autonomous Run Governance",
+      "id": "parallel-autonomous-run-governance",
+      "version": "0.2.3",
+      "description": "Coordinates isolated autonomous Spec Kit campaigns with bounded concurrency, mixed agents, resumable consolidation, governed post-merge closeout, schema 1.2, and an optional current intake-review gate before worker scheduling.",
+      "author": "Thorsten Hindermann",
+      "repository": "https://github.com/hindermath/spec-kit-preset-parallel-autonomous-run-governance",
+      "download_url": "https://github.com/hindermath/spec-kit-preset-parallel-autonomous-run-governance/archive/refs/tags/v0.2.3.zip",
+      "homepage": "https://github.com/hindermath/spec-kit-preset-parallel-autonomous-run-governance",
+      "documentation": "https://github.com/hindermath/spec-kit-preset-parallel-autonomous-run-governance/blob/v0.2.3/README.md",
+      "license": "MIT",
+      "requires": {
+        "speckit_version": ">=0.8.3"
+      },
+      "provides": {
+        "templates": 9,
+        "commands": 5,
+        "scripts": 2
+      },
+      "tags": [
+        "parallel",
+        "autonomous",
+        "governance",
+        "orchestration",
+        "resume",
+        "intake-review"
+      ],
+      "created_at": "2026-07-22T00:00:00Z",
+      "updated_at": "2026-07-22T00:00:00Z"
+    },
     "pirate": {
       "name": "Pirate Speak (Full)",
       "id": "pirate",
@@ -509,7 +598,7 @@
       "name": "Screenwriting",
       "id": "screenwriting",
       "version": "1.0.0",
-      "description": "Spec-Driven Development for screenwriting/scriptwriting/tutorials: feature films, television (pilot, episode, limited series), and stage plays. Adapts the Spec Kit workflow to screenplay craft — slug lines, action lines, act breaks, beat sheets, and industry-standard pitch documents replace prose fiction conventions. Supports three-act, Save the Cat, TV pilot, network episode, cable/streaming episode, and stage-play structural frameworks.",
+      "description": "Spec-Driven Development for screenwriting/scriptwriting/tutorials: feature films, television (pilot, episode, limited series), and stage plays. Adapts the Spec Kit workflow to screenplay craft \u2014 slug lines, action lines, act breaks, beat sheets, and industry-standard pitch documents replace prose fiction conventions. Supports three-act, Save the Cat, TV pilot, network episode, cable/streaming episode, and stage-play structural frameworks.",
       "author": "Andreas Daumann",
       "repository": "https://github.com/adaumann/speckit-preset-screenwriting",
       "download_url": "https://github.com/adaumann/speckit-preset-screenwriting/archive/refs/tags/v1.0.0.zip",
@@ -624,7 +713,7 @@
       "name": "Spec2Cloud",
       "id": "spec2cloud",
       "version": "1.1.0",
-      "description": "Spec-driven workflow tuned for shipping to Azure: spec → plan → tasks → implement → deploy.",
+      "description": "Spec-driven workflow tuned for shipping to Azure: spec \u2192 plan \u2192 tasks \u2192 implement \u2192 deploy.",
       "author": "Azure Samples",
       "repository": "https://github.com/Azure-Samples/Spec2Cloud",
       "download_url": "https://github.com/Azure-Samples/Spec2Cloud/releases/download/spec-kit-spec2cloud-v1.1.0/preset.zip",
@@ -652,7 +741,7 @@
       "id": "test-first-governance",
       "version": "1.3.0",
       "description": "Governs TDD with coverage-complete BDD/ATDD Gherkin scenarios, explicit suite ownership, professional test reports, traceability, and risk-based quality gates.",
-      "author": "Zoltán Katona, PhD",
+      "author": "Zolt\u00e1n Katona, PhD",
       "repository": "https://github.com/ka-zo/spec-kit-preset-test-first-governance",
       "download_url": "https://github.com/ka-zo/spec-kit-preset-test-first-governance/archive/refs/tags/1.3.0.zip",
       "homepage": "https://github.com/ka-zo/spec-kit-preset-test-first-governance",

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "specify-cli"
-version = "0.12.19.dev0"
+version = "0.13.4"
 description = "Specify CLI, part of GitHub Spec Kit. A tool to bootstrap your projects for Spec-Driven Development (SDD)."
 readme = "README.md"
 requires-python = ">=3.11"
@@ -48,6 +48,8 @@ packages = ["src/specify_cli"]
 "workflows/speckit" = "specify_cli/core_pack/workflows/speckit"
 # Bundled presets (installable via `specify preset add <name>` or `specify init --preset <name>`)
 "presets/lean" = "specify_cli/core_pack/presets/lean"
+# Community bundle catalog snapshot (used for offline discovery)
+"bundles/catalog.community.json" = "specify_cli/core_pack/bundles/catalog.community.json"
 
 [project.optional-dependencies]
 test = [

scripts/bash/create-new-feature.sh

@@ -90,6 +90,19 @@ if [ -z "$FEATURE_DESCRIPTION" ]; then
     exit 1
 fi
 
+MAX_FEATURE_NUMBER=9223372036854775807
+
+is_feature_number_in_range() {
+    local value="$1"
+    local normalized="${value#"${value%%[!0]*}"}"
+    [ -n "$normalized" ] || normalized=0
+    [ ${#normalized} -lt ${#MAX_FEATURE_NUMBER} ] && return 0
+    [ ${#normalized} -gt ${#MAX_FEATURE_NUMBER} ] && return 1
+    # Equal-length digit strings must be compared without arithmetic overflow.
+    # shellcheck disable=SC2071
+    [[ "$normalized" < "$MAX_FEATURE_NUMBER" || "$normalized" == "$MAX_FEATURE_NUMBER" ]]
+}
+
 # Function to get highest number from specs directory
 get_highest_from_specs() {
     local specs_dir="$1"
@@ -102,9 +115,11 @@ get_highest_from_specs() {
             # Match sequential prefixes (>=3 digits), but skip timestamp dirs.
             if echo "$dirname" | grep -Eq '^[0-9]{3,}-' && ! echo "$dirname" | grep -Eq '^[0-9]{8}-[0-9]{6}-'; then
                 number=$(echo "$dirname" | grep -Eo '^[0-9]+')
-                number=$((10#$number))
-                if [ "$number" -gt "$highest" ]; then
-                    highest=$number
+                if is_feature_number_in_range "$number"; then
+                    number=$((10#$number))
+                    if [ "$number" -gt "$highest" ]; then
+                        highest=$number
+                    fi
                 fi
             fi
         done
@@ -119,6 +134,19 @@ clean_branch_name() {
     echo "$name" | tr '[:upper:]' '[:lower:]' | sed 's/[^a-z0-9]/-/g' | sed 's/-\+/-/g' | sed 's/^-//' | sed 's/-$//'
 }
 
+# Quote a value for POSIX shell reuse, byte-identical to Python's shlex.quote
+# so the persistence hints match the Python variant exactly (printf %q output
+# differs between bash versions and from shlex.quote for spaces/metachars).
+shell_quote() {
+    local value="$1" LC_ALL=C
+    if [[ "$value" =~ ^[A-Za-z0-9_@%+=:,./-]+$ ]]; then
+        printf '%s' "$value"
+    else
+        local q="'\"'\"'"
+        printf "'%s'" "${value//\'/$q}"
+    fi
+}
+
 # Resolve repository root using common.sh functions which prioritize .specify
 SCRIPT_DIR="$(CDPATH="" cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
 source "$SCRIPT_DIR/common.sh"
@@ -202,9 +230,24 @@ if [ "$USE_TIMESTAMP" = true ]; then
     FEATURE_NUM=$(date +%Y%m%d-%H%M%S)
     BRANCH_NAME="${FEATURE_NUM}-${BRANCH_SUFFIX}"
 else
+    if [ -n "$BRANCH_NUMBER" ] && [[ ! "$BRANCH_NUMBER" =~ ^[0-9]+$ ]]; then
+        echo "Error: --number must be an unsigned integer, got '$BRANCH_NUMBER'" >&2
+        exit 1
+    fi
+
+    # Bash arithmetic is signed 64-bit; reject digit strings that would wrap.
+    if [ -n "$BRANCH_NUMBER" ] && ! is_feature_number_in_range "$BRANCH_NUMBER"; then
+        echo "Error: --number must be between 0 and $MAX_FEATURE_NUMBER, got '$BRANCH_NUMBER'" >&2
+        exit 1
+    fi
+
     # Determine branch number from existing feature directories
     if [ -z "$BRANCH_NUMBER" ]; then
         HIGHEST=$(get_highest_from_specs "$SPECS_DIR")
+        if [ "$HIGHEST" -eq "$MAX_FEATURE_NUMBER" ]; then
+            echo "Error: feature number must be between 0 and $MAX_FEATURE_NUMBER, got '9223372036854775808'" >&2
+            exit 1
+        fi
         BRANCH_NUMBER=$((HIGHEST + 1))
     fi
 
@@ -264,8 +307,8 @@ if [ "$DRY_RUN" != true ]; then
     _persist_feature_json "$REPO_ROOT" "$FEATURE_DIR"
 
     # Inform the user how to set feature state in their own shell
-    printf '# To persist: export SPECIFY_FEATURE=%q\n' "$BRANCH_NAME" >&2
-    printf '#              export SPECIFY_FEATURE_DIRECTORY=%q\n' "$FEATURE_DIR" >&2
+    printf '# To persist: export SPECIFY_FEATURE=%s\n' "$(shell_quote "$BRANCH_NAME")" >&2
+    printf '#              export SPECIFY_FEATURE_DIRECTORY=%s\n' "$(shell_quote "$FEATURE_DIR")" >&2
 fi
 
 if $JSON_MODE; then
@@ -295,7 +338,7 @@ else
     echo "SPEC_FILE: $SPEC_FILE"
     echo "FEATURE_NUM: $FEATURE_NUM"
     if [ "$DRY_RUN" != true ]; then
-        printf '# To persist in your shell: export SPECIFY_FEATURE=%q\n' "$BRANCH_NAME"
-        printf '#                           export SPECIFY_FEATURE_DIRECTORY=%q\n' "$FEATURE_DIR"
+        printf '# To persist in your shell: export SPECIFY_FEATURE=%s\n' "$(shell_quote "$BRANCH_NAME")"
+        printf '#                           export SPECIFY_FEATURE_DIRECTORY=%s\n' "$(shell_quote "$FEATURE_DIR")"
     fi
 fi

scripts/powershell/common.ps1

@@ -29,13 +29,16 @@ function Find-SpecifyRoot {
 # command against a member project from a monorepo root without cd.
 #
 # Precondition: $env:SPECIFY_INIT_DIR is set. Returns the validated project root,
-# or writes an error and exits 1. Strict by design: the path must exist and
+# or writes an error and exits 1 unless -ReturnNullOnError is set. Strict by
+# design: the path must exist and
 # contain .specify/, with no silent fallback. (An empty string is falsy, so the
 # caller's `if ($env:SPECIFY_INIT_DIR)` guard treats empty as unset.)
 #
 # This is the single resolver: bundled extensions inherit it by sourcing core
 # (e.g. the git extension's create-new-feature-branch) rather than duplicating it.
 function Resolve-SpecifyInitDir {
+    param([switch]$ReturnNullOnError)
+
     $initDir = $env:SPECIFY_INIT_DIR
     # Normalize: relative paths resolve against the current directory.
     if (-not [System.IO.Path]::IsPathRooted($initDir)) {
@@ -47,6 +50,7 @@ function Resolve-SpecifyInitDir {
     # "not a Spec Kit project" error below.
     if (-not $resolved -or -not (Test-Path -LiteralPath $resolved.Path -PathType Container)) {
         [Console]::Error.WriteLine("ERROR: SPECIFY_INIT_DIR does not point to an existing directory: $($env:SPECIFY_INIT_DIR)")
+        if ($ReturnNullOnError) { return $null }
         exit 1
     }
     # Resolve-Path echoes back any trailing separator from the input; trim it so
@@ -56,6 +60,7 @@ function Resolve-SpecifyInitDir {
     $initRoot = [System.IO.Path]::TrimEndingDirectorySeparator($resolved.Path)
     if (-not (Test-Path -LiteralPath (Join-Path $initRoot '.specify') -PathType Container)) {
         [Console]::Error.WriteLine("ERROR: SPECIFY_INIT_DIR is not a Spec Kit project (no .specify/ directory): $initRoot")
+        if ($ReturnNullOnError) { return $null }
         exit 1
     }
     return $initRoot
@@ -64,9 +69,11 @@ function Resolve-SpecifyInitDir {
 # Get repository root, prioritizing .specify directory
 # This prevents using a parent repository when spec-kit is initialized in a subdirectory
 function Get-RepoRoot {
+    param([switch]$ReturnNullOnError)
+
     # Explicit project override wins (see Resolve-SpecifyInitDir).
     if ($env:SPECIFY_INIT_DIR) {
-        return (Resolve-SpecifyInitDir)
+        return (Resolve-SpecifyInitDir -ReturnNullOnError:$ReturnNullOnError)
     }
 
     # First, look for .specify directory (spec-kit's own marker)
@@ -147,10 +154,12 @@ function Get-FeaturePathsEnv {
     # so pure path resolution never writes .specify/feature.json, which would
     # dirty the working tree or overwrite a pinned value (issue #3025).
     param(
-        [switch]$NoPersist
+        [switch]$NoPersist,
+        [switch]$ReturnNullOnError
     )
 
-    $repoRoot = Get-RepoRoot
+    $repoRoot = Get-RepoRoot -ReturnNullOnError:$ReturnNullOnError
+    if (-not $repoRoot) { return $null }
     $currentBranch = Get-CurrentBranch
 
     # Resolve feature directory.  Priority:
@@ -174,7 +183,8 @@ function Get-FeaturePathsEnv {
         try {
             $featureConfig = $featureJsonRaw | ConvertFrom-Json
         } catch {
-            [Console]::Error.WriteLine("ERROR: Failed to parse .specify/feature.json: $_")
+            [Console]::Error.WriteLine("ERROR: Feature directory not found. Set SPECIFY_FEATURE_DIRECTORY or ensure .specify/feature.json contains feature_directory.")
+            if ($ReturnNullOnError) { return $null }
             exit 1
         }
         if ($featureConfig.feature_directory) {
@@ -185,10 +195,12 @@ function Get-FeaturePathsEnv {
             }
         } else {
             [Console]::Error.WriteLine("ERROR: Feature directory not found. Set SPECIFY_FEATURE_DIRECTORY or ensure .specify/feature.json contains feature_directory.")
+            if ($ReturnNullOnError) { return $null }
             exit 1
         }
     } else {
         [Console]::Error.WriteLine("ERROR: Feature directory not found. Set SPECIFY_FEATURE_DIRECTORY or run the specify command to create .specify/feature.json.")
+        if ($ReturnNullOnError) { return $null }
         exit 1
     }
 
@@ -334,30 +346,64 @@ function Resolve-Template {
     if (Test-Path $presetsDir) {
         $registryFile = Join-Path $presetsDir '.registry'
         $sortedPresets = @()
+        $registryParsed = $false
         if (Test-Path $registryFile) {
             try {
                 $registryData = Get-Content $registryFile -Raw | ConvertFrom-Json
-                $presets = $registryData.presets
-                if ($presets) {
-                    $sortedPresets = $presets.PSObject.Properties |
+                if ($null -eq $registryData -or $registryData -isnot [PSCustomObject]) {
+                    throw 'Registry root must be an object'
+                }
+                $presetsProperty = $registryData.PSObject.Properties['presets']
+                if ($presetsProperty) {
+                    $presets = $presetsProperty.Value
+                    if ($null -eq $presets -or $presets -isnot [PSCustomObject]) {
+                        throw 'Registry presets must be an object'
+                    }
+                    $presetEntries = @($presets.PSObject.Properties)
+                    $priorityFor = {
+                        param($Entry)
+                        if ($Entry.Value -is [PSCustomObject]) {
+                            $priorityProperty = $Entry.Value.PSObject.Properties['priority']
+                            if ($priorityProperty) { return $priorityProperty.Value }
+                        }
+                        return 10
+                    }
+                    if ($presetEntries.Count -gt 1) {
+                        $allNumeric = $true
+                        $allStrings = $true
+                        foreach ($entry in $presetEntries) {
+                            $priority = & $priorityFor $entry
+                            if ($null -eq $priority -or $priority -isnot [ValueType]) {
+                                $allNumeric = $false
+                            }
+                            if ($null -eq $priority -or $priority -isnot [string]) {
+                                $allStrings = $false
+                            }
+                        }
+                        if (-not $allNumeric -and -not $allStrings) {
+                            throw 'Registry priorities are not mutually orderable'
+                        }
+                    }
+                    $sortedPresets = $presetEntries |
+                        Where-Object { $_.Value -is [PSCustomObject] } |
                         Where-Object { $null -eq $_.Value.enabled -or $_.Value.enabled -ne $false } |
-                        Sort-Object { if ($null -ne $_.Value.priority) { $_.Value.priority } else { 10 } } |
+                        Sort-Object { & $priorityFor $_ } |
                         ForEach-Object { $_.Name }
                 }
+                $registryParsed = $true
             } catch {
-                # Fallback: alphabetical directory order
-                $sortedPresets = @()
+                $registryParsed = $false
             }
         }
 
-        if ($sortedPresets.Count -gt 0) {
+        if ($registryParsed) {
             foreach ($presetId in $sortedPresets) {
                 $candidate = Join-Path $presetsDir "$presetId/templates/$TemplateName.md"
                 if (Test-Path $candidate) { return $candidate }
             }
         } else {
             # Fallback: alphabetical directory order
-            foreach ($preset in Get-ChildItem -Path $presetsDir -Directory -ErrorAction SilentlyContinue | Where-Object { $_.Name -notlike '.*' }) {
+            foreach ($preset in Get-ChildItem -Path $presetsDir -Directory -ErrorAction SilentlyContinue | Where-Object { $_.Name -notlike '.*' } | Sort-Object Name) {
                 $candidate = Join-Path $preset.FullName "templates/$TemplateName.md"
                 if (Test-Path $candidate) { return $candidate }
             }

scripts/powershell/create-new-feature.ps1

@@ -7,7 +7,7 @@ param(
     [switch]$DryRun,
     [string]$ShortName,
     [Parameter()]
-    [long]$Number = 0,
+    [string]$Number = '',
     [switch]$Timestamp,
     [switch]$Help,
     [Parameter(Position = 0, ValueFromRemainingArguments = $true)]
@@ -142,12 +142,13 @@ if ($ShortName) {
     $branchSuffix = Get-BranchName -Description $featureDesc
 }
 
-# Warn if -Number and -Timestamp are both specified. Use ContainsKey (not
-# `-ne 0`) so an explicit `-Number 0` is also detected, matching the bash twin's
-# `[ -n "$BRANCH_NUMBER" ]` check.
-if ($Timestamp -and $PSBoundParameters.ContainsKey('Number')) {
-    Write-Warning "[specify] Warning: -Number is ignored when -Timestamp is used"
-    $Number = 0
+# Treat an explicit empty string as omitted, matching the bash and Python twins.
+$hasNumber = $PSBoundParameters.ContainsKey('Number') -and $Number -ne ''
+
+# Warn if -Number and -Timestamp are both specified.
+if ($Timestamp -and $hasNumber) {
+    [Console]::Error.WriteLine("[specify] Warning: -Number is ignored when -Timestamp is used")
+    $Number = ''
 }
 
 # Determine branch prefix
@@ -158,11 +159,23 @@ if ($Timestamp) {
     # Determine branch number from existing feature directories. Auto-detect only
     # when -Number was not supplied; an explicit value (including 0) is honored,
     # matching the bash twin's `[ -z "$BRANCH_NUMBER" ]` check.
-    if (-not $PSBoundParameters.ContainsKey('Number')) {
-        $Number = (Get-HighestNumberFromSpecs -SpecsDir $specsDir) + 1
+    [long]$resolvedNumber = 0
+    if (-not $hasNumber) {
+        $highestNumber = Get-HighestNumberFromSpecs -SpecsDir $specsDir
+        if ($highestNumber -eq [long]::MaxValue) {
+            Write-Error "Error: feature number must be between 0 and $([long]::MaxValue), got '9223372036854775808'"
+            exit 1
+        }
+        $resolvedNumber = $highestNumber + 1
+    } elseif ($Number -notmatch '^[0-9]+$') {
+        Write-Error "Error: -Number must be an unsigned integer, got '$Number'"
+        exit 1
+    } elseif (-not [long]::TryParse($Number, [ref]$resolvedNumber)) {
+        Write-Error "Error: -Number must be between 0 and $([long]::MaxValue), got '$Number'"
+        exit 1
     }
 
-    $featureNum = ('{0:000}' -f $Number)
+    $featureNum = ('{0:000}' -f $resolvedNumber)
     $branchName = "$featureNum-$branchSuffix"
 }
 
@@ -183,9 +196,9 @@ if ($branchName.Length -gt $maxBranchLength) {
     $originalBranchName = $branchName
     $branchName = "$featureNum-$truncatedSuffix"
 
-    Write-Warning "[specify] Branch name exceeded GitHub's 244-byte limit"
-    Write-Warning "[specify] Original: $originalBranchName ($($originalBranchName.Length) bytes)"
-    Write-Warning "[specify] Truncated to: $branchName ($($branchName.Length) bytes)"
+    [Console]::Error.WriteLine("[specify] Warning: Branch name exceeded GitHub's 244-byte limit")
+    [Console]::Error.WriteLine("[specify] Original: $originalBranchName ($($originalBranchName.Length) bytes)")
+    [Console]::Error.WriteLine("[specify] Truncated to: $branchName ($($branchName.Length) bytes)")
 }
 
 $featureDir = Join-Path $specsDir $branchName
@@ -225,6 +238,13 @@ if (-not $DryRun) {
     # Set environment variables for the current session
     $env:SPECIFY_FEATURE = $branchName
     $env:SPECIFY_FEATURE_DIRECTORY = $featureDir
+
+    $quotedBranchName = "'" + $branchName.Replace("'", "''") + "'"
+    $quotedFeatureDir = "'" + $featureDir.Replace("'", "''") + "'"
+    $featureAssignment = '$env:SPECIFY_FEATURE = ' + $quotedBranchName
+    $directoryAssignment = '$env:SPECIFY_FEATURE_DIRECTORY = ' + $quotedFeatureDir
+    [Console]::Error.WriteLine("# To persist: $featureAssignment")
+    [Console]::Error.WriteLine("#              $directoryAssignment")
 }
 
 if ($Json) {
@@ -242,7 +262,7 @@ if ($Json) {
     Write-Output "SPEC_FILE: $specFile"
     Write-Output "FEATURE_NUM: $featureNum"
     if (-not $DryRun) {
-        Write-Output "SPECIFY_FEATURE set to: $branchName"
-        Write-Output "SPECIFY_FEATURE_DIRECTORY set to: $featureDir"
+        Write-Output "# To persist in your shell: $featureAssignment"
+        Write-Output "#                           $directoryAssignment"
     }
 }

scripts/powershell/setup-plan.ps1

@@ -4,7 +4,10 @@
 [CmdletBinding()]
 param(
     [switch]$Json,
-    [switch]$Help
+    [switch]$Help,
+    # Capture extra positional arguments to match Bash/Python behavior.
+    [Parameter(ValueFromRemainingArguments = $true)]
+    [string[]]$RemainingArgs
 )
 
 $ErrorActionPreference = 'Stop'
@@ -21,7 +24,11 @@ if ($Help) {
 . "$PSScriptRoot/common.ps1"
 
 # Get all paths and variables from common functions
-$paths = Get-FeaturePathsEnv
+$paths = Get-FeaturePathsEnv -ReturnNullOnError
+if (-not $paths) {
+    [Console]::Error.WriteLine("ERROR: Failed to resolve feature paths")
+    exit 1
+}
 
 # Ensure the feature directory exists
 New-Item -ItemType Directory -Path $paths.FEATURE_DIR -Force | Out-Null

scripts/powershell/setup-tasks.ps1

@@ -3,21 +3,34 @@
 [CmdletBinding()]
 param(
     [switch]$Json,
-    [switch]$Help
+    [switch]$Help,
+    [Parameter(ValueFromRemainingArguments = $true)]
+    [string[]]$RemainingArgs
 )
 
 $ErrorActionPreference = 'Stop'
 
+# Help wins over unknown-argument validation to match the Bash/Python
+# variants, which stop at --help and exit 0.
 if ($Help) {
     Write-Output "Usage: setup-tasks.ps1 [-Json] [-Help]"
     exit 0
 }
 
+if ($RemainingArgs.Count -gt 0) {
+    [Console]::Error.WriteLine("ERROR: Unknown option '$($RemainingArgs[0])'")
+    exit 1
+}
+
 # Source common functions
 . "$PSScriptRoot/common.ps1"
 
 # Get feature paths
-$paths = Get-FeaturePathsEnv
+$paths = Get-FeaturePathsEnv -ReturnNullOnError
+if (-not $paths) {
+    [Console]::Error.WriteLine("ERROR: Failed to resolve feature paths")
+    exit 1
+}
 
 if (-not (Test-Path $paths.IMPL_PLAN -PathType Leaf)) {
     [Console]::Error.WriteLine("ERROR: plan.md not found in $($paths.FEATURE_DIR)")
@@ -45,8 +58,8 @@ if (Test-Path $paths.QUICKSTART) { $docs += 'quickstart.md' }
 # Resolve tasks template through override stack
 $tasksTemplate = Resolve-Template -TemplateName 'tasks-template' -RepoRoot $paths.REPO_ROOT
 if (-not $tasksTemplate -or -not (Test-Path -LiteralPath $tasksTemplate -PathType Leaf)) {
-    $expectedCoreTemplate = Join-Path $paths.REPO_ROOT '.specify/templates/tasks-template.md'
-    [Console]::Error.WriteLine("ERROR: Tasks template not found for repository root: $($paths.REPO_ROOT)`nTemplate resolution order: overrides -> presets -> extensions -> core.`nExpected shared/core template location: $expectedCoreTemplate`nTo continue, verify whether 'tasks-template.md' is available in '.specify/templates/overrides/', preset templates, extension templates, or restore the shared/core templates (for example by re-running 'specify init') so that '.specify/templates/tasks-template.md' exists.")
+    [Console]::Error.WriteLine("ERROR: Could not resolve required tasks-template from the template override stack for $($paths.REPO_ROOT)")
+    [Console]::Error.WriteLine("Template 'tasks-template' was not found in any supported location (overrides, presets, extensions, or shared core). Add an override at .specify/templates/overrides/tasks-template.md, or run 'specify init' / reinstall shared infra to restore the core .specify/templates/tasks-template.md template.")
     exit 1
 }
 $tasksTemplate = (Resolve-Path -LiteralPath $tasksTemplate).Path

scripts/python/common.py

@@ -84,7 +84,7 @@ def read_feature_json_feature_directory(repo_root: Path) -> str:
         return ""
     try:
         data = json.loads(feature_json.read_text(encoding="utf-8"))
-    except (OSError, json.JSONDecodeError):
+    except (OSError, UnicodeError, json.JSONDecodeError):
         return ""
     value = data.get("feature_directory") if isinstance(data, dict) else None
     return value if isinstance(value, str) else ""
@@ -95,16 +95,17 @@ def _json_dump(data: dict[str, str]) -> str:
 
 
 def persist_feature_json(repo_root: Path, feature_dir_value: str) -> None:
+    # Strip the repo root prefix lexically (no resolve()) to mirror the
+    # Bash/PowerShell helpers: with a symlinked <repo>/specs, resolve() would
+    # escape the repo and persist a machine-specific absolute path instead of
+    # the relative "specs/NNN-name" the other variants store.
     value = feature_dir_value
-    try:
-        relative = Path(value)
-        if relative.is_absolute():
-            try:
-                value = relative.resolve().relative_to(repo_root.resolve()).as_posix()
-            except ValueError:
-                value = str(relative)
-    except OSError:
-        pass
+    relative = Path(value)
+    if relative.is_absolute():
+        try:
+            value = relative.relative_to(repo_root).as_posix()
+        except ValueError:
+            value = str(relative)
 
     current = read_feature_json_feature_directory(repo_root)
     if current == value:
@@ -112,9 +113,8 @@ def persist_feature_json(repo_root: Path, feature_dir_value: str) -> None:
 
     specify_dir = repo_root / ".specify"
     specify_dir.mkdir(parents=True, exist_ok=True)
-    (specify_dir / "feature.json").write_text(
-        _json_dump({"feature_directory": value}),
-        encoding="utf-8",
+    (specify_dir / "feature.json").write_bytes(
+        _json_dump({"feature_directory": value}).encode("utf-8")
     )
 
 
@@ -182,6 +182,78 @@ def get_feature_paths(
     )
 
 
+def _sorted_preset_ids(presets_dir: Path) -> list[str]:
+    registry = presets_dir / ".registry"
+    if registry.is_file():
+        # Mirrors bash: any failure while reading or sorting the registry
+        # (invalid JSON, non-dict shapes, unorderable priority values) falls
+        # back to the directory scan below.
+        try:
+            data = json.loads(registry.read_text(encoding="utf-8"))
+            presets = data.get("presets", {})
+            return [
+                pid
+                for pid, meta in sorted(
+                    presets.items(),
+                    key=lambda kv: kv[1].get("priority", 10)
+                    if isinstance(kv[1], dict)
+                    else 10,
+                )
+                if isinstance(meta, dict) and meta.get("enabled", True) is not False
+            ]
+        except Exception:
+            pass
+    try:
+        return sorted(
+            p.name
+            for p in presets_dir.iterdir()
+            if p.is_dir() and not p.name.startswith(".")
+        )
+    except OSError:
+        return []
+
+
+def resolve_template(template_name: str, repo_root: Path) -> Path | None:
+    """Resolve a template name to a file path using the priority stack.
+
+    Order (mirrors resolve_template in scripts/bash/common.sh):
+      1. .specify/templates/overrides/
+      2. .specify/presets/<preset-id>/templates/ (sorted by .registry priority)
+      3. .specify/extensions/<ext-id>/templates/ (hidden directories skipped)
+      4. .specify/templates/ (core)
+    """
+    base = repo_root / ".specify" / "templates"
+
+    override = base / "overrides" / f"{template_name}.md"
+    if override.is_file():
+        return override
+
+    presets_dir = repo_root / ".specify" / "presets"
+    if presets_dir.is_dir():
+        for preset_id in _sorted_preset_ids(presets_dir):
+            candidate = presets_dir / preset_id / "templates" / f"{template_name}.md"
+            if candidate.is_file():
+                return candidate
+
+    ext_dir = repo_root / ".specify" / "extensions"
+    if ext_dir.is_dir():
+        try:
+            extensions = sorted(p for p in ext_dir.iterdir() if p.is_dir())
+        except OSError:
+            extensions = []
+        for ext in extensions:
+            if ext.name.startswith("."):
+                continue
+            candidate = ext / "templates" / f"{template_name}.md"
+            if candidate.is_file():
+                return candidate
+
+    core = base / f"{template_name}.md"
+    if core.is_file():
+        return core
+    return None
+
+
 def get_invoke_separator(repo_root: Path) -> str:
     integration_json = repo_root / ".specify" / "integration.json"
     if not integration_json.is_file():

scripts/python/create_new_feature.py

@@ -0,0 +1,355 @@
+#!/usr/bin/env python3
+"""Create a new feature directory and spec file."""
+
+from __future__ import annotations
+
+import datetime
+import json
+import re
+import shlex
+import shutil
+import sys
+from dataclasses import dataclass
+from pathlib import Path
+
+try:
+    from common import get_repo_root, persist_feature_json, resolve_template
+except ImportError:  # pragma: no cover - direct execution from unusual cwd
+    sys.path.insert(0, str(Path(__file__).resolve().parent))
+    from common import get_repo_root, persist_feature_json, resolve_template
+
+
+def _json_line(payload: object) -> str:
+    return json.dumps(payload, ensure_ascii=False, separators=(",", ":")) + "\n"
+
+
+_STOP_WORDS = frozenset(
+    """
+    i a an the to for of in on at by with from is are was were be been being
+    have has had do does did will would should could can may might must shall
+    this that these those my your our their want need add get set
+    """.split()
+)
+
+_MAX_BRANCH_LENGTH = 244
+_MAX_FEATURE_NUMBER = 2**63 - 1
+
+
+def _int64_from_digits(value: str) -> int | None:
+    normalized = value.lstrip("0") or "0"
+    maximum = str(_MAX_FEATURE_NUMBER)
+    if len(normalized) > len(maximum) or (
+        len(normalized) == len(maximum) and normalized > maximum
+    ):
+        return None
+    return int(normalized, 10)
+
+
+def _persistence_assignments(
+    branch_name: str, feature_dir: str, *, powershell: bool
+) -> tuple[str, str]:
+    if powershell:
+        quoted_branch = "'" + branch_name.replace("'", "''") + "'"
+        quoted_dir = "'" + feature_dir.replace("'", "''") + "'"
+        return (
+            f"$env:SPECIFY_FEATURE = {quoted_branch}",
+            f"$env:SPECIFY_FEATURE_DIRECTORY = {quoted_dir}",
+        )
+    return (
+        f"export SPECIFY_FEATURE={shlex.quote(branch_name)}",
+        f"export SPECIFY_FEATURE_DIRECTORY={shlex.quote(feature_dir)}",
+    )
+
+
+def _usage(argv0: str) -> str:
+    return (
+        f"Usage: {argv0} [--json] [--dry-run] [--allow-existing-branch] "
+        "[--short-name <name>] [--number N] [--timestamp] <feature_description>"
+    )
+
+
+def _help_text(argv0: str) -> str:
+    return f"""{_usage(argv0)}
+
+Options:
+  --json              Output in JSON format
+  --dry-run           Compute feature name and paths without creating directories or files
+  --allow-existing-branch  Reuse an existing feature directory if it already exists
+  --short-name <name> Provide a custom short name (2-4 words) for the feature
+  --number N          Specify branch number manually (overrides auto-detection)
+  --timestamp         Use timestamp prefix (YYYYMMDD-HHMMSS) instead of sequential numbering
+  --help, -h          Show this help message
+
+Examples:
+  {argv0} 'Add user authentication system' --short-name 'user-auth'
+  {argv0} 'Implement OAuth2 integration for API' --number 5
+  {argv0} --timestamp --short-name 'user-auth' 'Add user authentication'
+"""
+
+
+@dataclass(frozen=True)
+class Args:
+    json_mode: bool = False
+    dry_run: bool = False
+    allow_existing: bool = False
+    short_name: str = ""
+    branch_number: str = ""
+    use_timestamp: bool = False
+    description: str = ""
+
+
+def _parse_args(argv: list[str], argv0: str) -> Args:
+    json_mode = False
+    dry_run = False
+    allow_existing = False
+    short_name = ""
+    branch_number = ""
+    use_timestamp = False
+    rest: list[str] = []
+
+    i = 0
+    while i < len(argv):
+        arg = argv[i]
+        if arg == "--json":
+            json_mode = True
+        elif arg == "--dry-run":
+            dry_run = True
+        elif arg == "--allow-existing-branch":
+            allow_existing = True
+        elif arg in {"--short-name", "--number"}:
+            if i + 1 >= len(argv) or argv[i + 1].startswith("--"):
+                print(f"Error: {arg} requires a value", file=sys.stderr)
+                raise SystemExit(1)
+            i += 1
+            if arg == "--short-name":
+                short_name = argv[i]
+            else:
+                branch_number = argv[i]
+        elif arg == "--timestamp":
+            use_timestamp = True
+        elif arg in {"--help", "-h"}:
+            sys.stdout.write(_help_text(argv0))
+            raise SystemExit(0)
+        else:
+            rest.append(arg)
+        i += 1
+
+    description = " ".join(rest).strip()
+    if not description:
+        if rest:
+            print(
+                "Error: Feature description cannot be empty or contain only whitespace",
+                file=sys.stderr,
+            )
+        else:
+            print(_usage(argv0), file=sys.stderr)
+        raise SystemExit(1)
+
+    return Args(
+        json_mode=json_mode,
+        dry_run=dry_run,
+        allow_existing=allow_existing,
+        short_name=short_name,
+        branch_number=branch_number,
+        use_timestamp=use_timestamp,
+        description=description,
+    )
+
+
+def _clean_branch_name(name: str) -> str:
+    cleaned = re.sub(r"[^a-z0-9]", "-", name.lower())
+    cleaned = re.sub(r"-+", "-", cleaned)
+    return cleaned.strip("-")
+
+
+def _generate_branch_name(description: str) -> str:
+    clean = re.sub(r"[^a-z0-9]", " ", description.lower())
+    meaningful: list[str] = []
+    for word in clean.split():
+        if word in _STOP_WORDS:
+            continue
+        if len(word) >= 3:
+            meaningful.append(word)
+        # Keep short words that appear as an uppercase acronym in the original,
+        # mirroring the bash twin's case-sensitive `grep -qw` check.
+        elif re.search(
+            rf"(?<![0-9A-Za-z_]){re.escape(word.upper())}(?![0-9A-Za-z_])",
+            description,
+        ):
+            meaningful.append(word)
+
+    if meaningful:
+        max_words = 4 if len(meaningful) == 4 else 3
+        return "-".join(meaningful[:max_words])
+
+    cleaned = _clean_branch_name(description)
+    return "-".join([part for part in cleaned.split("-") if part][:3])
+
+
+def _get_highest_from_specs(specs_dir: Path) -> int:
+    highest = 0
+    if not specs_dir.is_dir():
+        return highest
+    for entry in specs_dir.iterdir():
+        if not entry.is_dir():
+            continue
+        name = entry.name
+        # Match sequential prefixes (>=3 digits), but skip timestamp dirs.
+        if re.match(r"^[0-9]{3,}-", name) and not re.match(
+            r"^[0-9]{8}-[0-9]{6}-", name
+        ):
+            number = _int64_from_digits(re.match(r"^[0-9]+", name).group())
+            if number is not None:
+                highest = max(highest, number)
+    return highest
+
+
+def main(argv: list[str] | None = None) -> int:
+    argv0 = sys.argv[0]
+    args = _parse_args(list(argv if argv is not None else sys.argv[1:]), argv0)
+
+    repo_root = get_repo_root(Path(__file__))
+    specs_dir = repo_root / "specs"
+    if not args.dry_run:
+        specs_dir.mkdir(parents=True, exist_ok=True)
+
+    if args.short_name:
+        branch_suffix = _clean_branch_name(args.short_name)
+    else:
+        branch_suffix = _generate_branch_name(args.description)
+
+    branch_number = args.branch_number
+    if args.use_timestamp and branch_number:
+        print(
+            "[specify] Warning: --number is ignored when --timestamp is used",
+            file=sys.stderr,
+        )
+        branch_number = ""
+
+    if args.use_timestamp:
+        feature_num = datetime.datetime.now().strftime("%Y%m%d-%H%M%S")
+    else:
+        if branch_number:
+            # Mirrors bash: $((10#$BRANCH_NUMBER)) only accepts unsigned
+            # decimal digits, rejecting signs, whitespace, and other
+            # characters that int() would otherwise tolerate.
+            if not re.fullmatch(r"[0-9]+", branch_number):
+                print(
+                    "Error: --number must be an unsigned integer, "
+                    f"got '{branch_number}'",
+                    file=sys.stderr,
+                )
+                return 1
+            number = _int64_from_digits(branch_number)
+            if number is None:
+                print(
+                    "Error: --number must be between 0 and "
+                    f"{_MAX_FEATURE_NUMBER}, got '{branch_number}'",
+                    file=sys.stderr,
+                )
+                return 1
+        else:
+            number = _get_highest_from_specs(specs_dir) + 1
+        if number > _MAX_FEATURE_NUMBER:
+            rejected_number = branch_number or str(number)
+            number_label = "--number" if branch_number else "feature number"
+            print(
+                f"Error: {number_label} must be between 0 and "
+                f"{_MAX_FEATURE_NUMBER}, got '{rejected_number}'",
+                file=sys.stderr,
+            )
+            return 1
+        feature_num = f"{number:03d}"
+
+    max_suffix_length = _MAX_BRANCH_LENGTH - (len(feature_num) + 1)
+    if max_suffix_length <= 0:
+        print("Error: feature number is too long for a branch name", file=sys.stderr)
+        return 1
+
+    branch_name = f"{feature_num}-{branch_suffix}"
+
+    # GitHub enforces a 244-byte limit on branch names.
+    if len(branch_name) > _MAX_BRANCH_LENGTH:
+        truncated_suffix = re.sub(r"-$", "", branch_suffix[:max_suffix_length])
+        original_branch_name = branch_name
+        branch_name = f"{feature_num}-{truncated_suffix}"
+        print(
+            "[specify] Warning: Branch name exceeded GitHub's 244-byte limit",
+            file=sys.stderr,
+        )
+        print(
+            f"[specify] Original: {original_branch_name} "
+            f"({len(original_branch_name)} bytes)",
+            file=sys.stderr,
+        )
+        print(
+            f"[specify] Truncated to: {branch_name} ({len(branch_name)} bytes)",
+            file=sys.stderr,
+        )
+
+    feature_dir = specs_dir / branch_name
+    spec_file = feature_dir / "spec.md"
+
+    if not args.dry_run:
+        if feature_dir.is_dir() and not args.allow_existing:
+            if args.use_timestamp:
+                print(
+                    f"Error: Feature directory '{feature_dir}' already exists. "
+                    "Rerun to get a new timestamp or use a different --short-name.",
+                    file=sys.stderr,
+                )
+            else:
+                print(
+                    f"Error: Feature directory '{feature_dir}' already exists. "
+                    "Please use a different feature name or specify a different "
+                    "number with --number.",
+                    file=sys.stderr,
+                )
+            return 1
+
+        feature_dir.mkdir(parents=True, exist_ok=True)
+
+        if not spec_file.is_file():
+            template = resolve_template("spec-template", repo_root)
+            if template is not None and template.is_file():
+                shutil.copy(template, spec_file)
+            else:
+                print(
+                    "Warning: Spec template not found; created empty spec file",
+                    file=sys.stderr,
+                )
+                spec_file.touch()
+
+        # Persist to .specify/feature.json so downstream commands can find the feature.
+        persist_feature_json(repo_root, f"specs/{branch_name}")
+
+        # Inform the user how to set feature state in their own shell.
+        feature_assignment, directory_assignment = _persistence_assignments(
+            branch_name,
+            str(feature_dir),
+            powershell=sys.platform == "win32",
+        )
+        print(f"# To persist: {feature_assignment}", file=sys.stderr)
+        print(f"#              {directory_assignment}", file=sys.stderr)
+
+    if args.json_mode:
+        payload: dict[str, object] = {
+            "BRANCH_NAME": branch_name,
+            "SPEC_FILE": str(spec_file),
+            "FEATURE_NUM": feature_num,
+        }
+        if args.dry_run:
+            payload["DRY_RUN"] = True
+        sys.stdout.write(_json_line(payload))
+    else:
+        print(f"BRANCH_NAME: {branch_name}")
+        print(f"SPEC_FILE: {spec_file}")
+        print(f"FEATURE_NUM: {feature_num}")
+        if not args.dry_run:
+            print(f"# To persist in your shell: {feature_assignment}")
+            print(f"#                           {directory_assignment}")
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

scripts/python/setup_plan.py

@@ -0,0 +1,86 @@
+#!/usr/bin/env python3
+"""Setup implementation plan for a feature."""
+
+from __future__ import annotations
+
+import json
+import shutil
+import sys
+from pathlib import Path
+
+try:
+    from common import get_feature_paths, resolve_template
+except ImportError:  # pragma: no cover - direct execution from unusual cwd
+    sys.path.insert(0, str(Path(__file__).resolve().parent))
+    from common import get_feature_paths, resolve_template
+
+
+def _json_line(payload: object) -> str:
+    return json.dumps(payload, ensure_ascii=False, separators=(",", ":")) + "\n"
+
+
+def _help_text(argv0: str) -> str:
+    return f"""Usage: {argv0} [--json]
+  --json    Output results in JSON format
+  --help    Show this help message
+"""
+
+
+def main(argv: list[str] | None = None) -> int:
+    args = list(argv if argv is not None else sys.argv[1:])
+    json_mode = False
+    for arg in args:
+        if arg == "--json":
+            json_mode = True
+        elif arg in {"--help", "-h"}:
+            sys.stdout.write(_help_text(sys.argv[0]))
+            return 0
+        # Other arguments are accepted and silently ignored, matching setup-plan.sh.
+
+    try:
+        paths = get_feature_paths(script_file=Path(__file__))
+    except SystemExit as exc:
+        if exc.code == 0:
+            return 0
+        print("ERROR: Failed to resolve feature paths", file=sys.stderr)
+        return int(exc.code) if isinstance(exc.code, int) else 1
+
+    paths.feature_dir.mkdir(parents=True, exist_ok=True)
+
+    # Status messages go to stderr in JSON mode so stdout stays pure JSON.
+    status_stream = sys.stderr if json_mode else sys.stdout
+    if paths.impl_plan.is_file():
+        print(
+            f"Plan already exists at {paths.impl_plan}, skipping template copy",
+            file=status_stream,
+        )
+    else:
+        template = resolve_template("plan-template", paths.repo_root)
+        if template is not None and template.is_file():
+            shutil.copy(template, paths.impl_plan)
+            print(f"Copied plan template to {paths.impl_plan}", file=status_stream)
+        else:
+            print("Warning: Plan template not found", file=status_stream)
+            paths.impl_plan.touch()
+
+    if json_mode:
+        sys.stdout.write(
+            _json_line(
+                {
+                    "FEATURE_SPEC": str(paths.feature_spec),
+                    "IMPL_PLAN": str(paths.impl_plan),
+                    "SPECS_DIR": str(paths.feature_dir),
+                    "BRANCH": paths.current_branch,
+                }
+            )
+        )
+    else:
+        print(f"FEATURE_SPEC: {paths.feature_spec}")
+        print(f"IMPL_PLAN: {paths.impl_plan}")
+        print(f"SPECS_DIR: {paths.feature_dir}")
+        print(f"BRANCH: {paths.current_branch}")
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

scripts/python/setup_tasks.py

@@ -0,0 +1,145 @@
+#!/usr/bin/env python3
+"""Check tasks prerequisites and resolve the tasks template."""
+
+from __future__ import annotations
+
+import json
+import sys
+from pathlib import Path
+
+try:
+    from common import (
+        FeaturePaths,
+        format_speckit_command,
+        get_feature_paths,
+        resolve_template,
+    )
+except ImportError:  # pragma: no cover - direct execution from unusual cwd
+    sys.path.insert(0, str(Path(__file__).resolve().parent))
+    from common import (
+        FeaturePaths,
+        format_speckit_command,
+        get_feature_paths,
+        resolve_template,
+    )
+
+
+def _json_line(payload: object) -> str:
+    return json.dumps(payload, ensure_ascii=False, separators=(",", ":")) + "\n"
+
+
+def _help_text(argv0: str) -> str:
+    return f"""Usage: {argv0} [--json]
+  --json    Output results in JSON format
+  --help    Show this help message
+"""
+
+
+def _dir_has_entries(path: Path) -> bool:
+    try:
+        return path.is_dir() and any(path.iterdir())
+    except OSError:
+        return False
+
+
+def _available_docs(paths: FeaturePaths) -> list[str]:
+    docs: list[str] = []
+    if paths.research.is_file():
+        docs.append("research.md")
+    if paths.data_model.is_file():
+        docs.append("data-model.md")
+    if _dir_has_entries(paths.contracts_dir):
+        docs.append("contracts/")
+    if paths.quickstart.is_file():
+        docs.append("quickstart.md")
+    return docs
+
+
+def _check_file(path: Path, description: str) -> None:
+    marker = "✓" if path.is_file() else "✗"
+    print(f"  {marker} {description}")
+
+
+def _check_dir(path: Path, description: str) -> None:
+    marker = "✓" if _dir_has_entries(path) else "✗"
+    print(f"  {marker} {description}")
+
+
+def main(argv: list[str] | None = None) -> int:
+    json_mode = False
+    for arg in list(argv if argv is not None else sys.argv[1:]):
+        if arg == "--json":
+            json_mode = True
+        elif arg in {"--help", "-h"}:
+            sys.stdout.write(_help_text(sys.argv[0]))
+            return 0
+        else:
+            print(f"ERROR: Unknown option '{arg}'", file=sys.stderr)
+            return 1
+
+    try:
+        paths = get_feature_paths(script_file=Path(__file__))
+    except SystemExit as exc:
+        if exc.code == 0:
+            return 0
+        print("ERROR: Failed to resolve feature paths", file=sys.stderr)
+        return int(exc.code) if isinstance(exc.code, int) else 1
+
+    if not paths.impl_plan.is_file():
+        print(f"ERROR: plan.md not found in {paths.feature_dir}", file=sys.stderr)
+        print(
+            f"Run {format_speckit_command('plan', paths.repo_root)} first to create the implementation plan.",
+            file=sys.stderr,
+        )
+        return 1
+
+    if not paths.feature_spec.is_file():
+        print(f"ERROR: spec.md not found in {paths.feature_dir}", file=sys.stderr)
+        print(
+            f"Run {format_speckit_command('specify', paths.repo_root)} first to create the feature structure.",
+            file=sys.stderr,
+        )
+        return 1
+
+    docs = _available_docs(paths)
+
+    tasks_template = resolve_template("tasks-template", paths.repo_root)
+    if tasks_template is None or not tasks_template.is_file():
+        print(
+            "ERROR: Could not resolve required tasks-template from the template "
+            f"override stack for {paths.repo_root}",
+            file=sys.stderr,
+        )
+        print(
+            "Template 'tasks-template' was not found in any supported location "
+            "(overrides, presets, extensions, or shared core). Add an override at "
+            ".specify/templates/overrides/tasks-template.md, or run 'specify init' "
+            "/ reinstall shared infra to restore the core "
+            ".specify/templates/tasks-template.md template.",
+            file=sys.stderr,
+        )
+        return 1
+
+    if json_mode:
+        sys.stdout.write(
+            _json_line(
+                {
+                    "FEATURE_DIR": str(paths.feature_dir),
+                    "AVAILABLE_DOCS": docs,
+                    "TASKS_TEMPLATE": str(tasks_template),
+                }
+            )
+        )
+    else:
+        print(f"FEATURE_DIR: {paths.feature_dir}")
+        print(f"TASKS_TEMPLATE: {tasks_template}")
+        print("AVAILABLE_DOCS:")
+        _check_file(paths.research, "research.md")
+        _check_file(paths.data_model, "data-model.md")
+        _check_dir(paths.contracts_dir, "contracts/")
+        _check_file(paths.quickstart, "quickstart.md")
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

src/specify_cli/__init__.py

@@ -140,10 +140,9 @@ def _install_shared_infra(
     """Install shared infrastructure files into *project_path*.
 
     Copies ``.specify/scripts/<variant>/`` and ``.specify/templates/`` from
-    the bundled core_pack or source checkout, where ``<variant>`` is
-    ``bash`` when *script_type* is ``"sh"``, ``python`` when it is ``"py"``,
-    and ``powershell`` when it is ``"ps"``.  Tracks all installed files in
-    ``speckit.manifest.json``.
+    the bundled core_pack or source checkout. ``sh`` installs Bash, ``ps``
+    installs PowerShell, and ``py`` installs Python plus the platform shell
+    fallback. Tracks all installed files in ``speckit.manifest.json``.
 
     Shared scripts and page templates are processed to resolve
     ``__SPECKIT_COMMAND_<NAME>__`` placeholders using *invoke_separator*

src/specify_cli/_invocation_style.py

@@ -18,6 +18,7 @@ ALWAYS_SLASH_AGENTS: frozenset[str] = frozenset({"devin", "grok", "trae", "zed"}
 CONDITIONAL_SLASH_AGENTS: frozenset[str] = frozenset(
     {
         "agy",
+        "bob",
         "claude",
         "copilot",
         "cursor-agent",

src/specify_cli/agents.py

@@ -7,7 +7,6 @@ command files into agent-specific directories in the correct format.
 """
 
 import os
-import platform
 import re
 from copy import deepcopy
 from pathlib import Path
@@ -114,13 +113,24 @@ class CommandRegistrar:
         if not content.startswith("---"):
             return {}, content
 
-        # Find second ---
-        end_marker = content.find("---", 3)
-        if end_marker == -1:
+        # The closing delimiter is a line that is exactly ``---`` (a YAML
+        # document separator), not any ``---`` substring. Scanning with
+        # ``content.find("---", 3)`` stops at the first ``---`` *anywhere* —
+        # including one embedded in a frontmatter value (e.g. a description like
+        # "Separate sections with ---") or inside an indented literal block —
+        # which truncates the frontmatter and spills the remainder into the
+        # body. Match on line boundaries instead, mirroring the line-anchored
+        # scan in ``VibeIntegration._inject_frontmatter_flag``.
+        lines = content.splitlines(keepends=True)
+        end_line = next(
+            (i for i in range(1, len(lines)) if lines[i].rstrip() == "---"),
+            None,
+        )
+        if end_line is None:
             return {}, content
 
-        frontmatter_str = content[3:end_marker].strip()
-        body = content[end_marker + 3 :].strip()
+        frontmatter_str = "".join(lines[1:end_line]).strip()
+        body = "".join(lines[end_line + 1 :]).strip()
 
         try:
             frontmatter = yaml.safe_load(frontmatter_str) or {}
@@ -475,26 +485,19 @@ class CommandRegistrar:
             init_opts = {}
 
         script_variant = init_opts.get("script")
-        if script_variant not in {"sh", "ps"}:
-            fallback_order = []
-            default_variant = (
-                "ps" if platform.system().lower().startswith("win") else "sh"
+        if scripts:
+            from specify_cli.integrations.base import IntegrationBase
+
+            script_variant = IntegrationBase.select_script_variant(
+                script_variant, scripts
             )
-            secondary_variant = "sh" if default_variant == "ps" else "ps"
-
-            if default_variant in scripts:
-                fallback_order.append(default_variant)
-            if secondary_variant in scripts:
-                fallback_order.append(secondary_variant)
-
-            for key in scripts:
-                if key not in fallback_order:
-                    fallback_order.append(key)
-
-            script_variant = fallback_order[0] if fallback_order else None
 
         script_command = scripts.get(script_variant) if script_variant else None
         if script_command:
+            if script_variant == "py":
+                script_command = IntegrationBase.build_python_invocation(
+                    script_command, project_root
+                )
             script_command = script_command.replace("{ARGS}", "$ARGUMENTS")
             body = body.replace("{SCRIPT}", script_command)
 
@@ -637,6 +640,37 @@ class CommandRegistrar:
         is_cline_ext = agent_name == "cline" and source_id != "core"
         source_root = source_dir.resolve()
 
+        # Resolve the command-reference separator for the file THIS registrar
+        # is about to write.  The separator must match the *output layout* the
+        # registrar produces for this agent — not the project's persisted
+        # ``ai_skills`` flag, and not unrelated sibling directories on disk.  A
+        # skill scaffold ("/SKILL.md") uses the skills separator; any
+        # command-layout output (".md", ".agent.md", ".toml", …) uses the
+        # command separator.
+        #
+        # This holds for the *active* agent too.  Dual-layout agents (Bob,
+        # Copilot) write their skills via their own setup()/skills path, so
+        # ``register_commands`` only ever emits their command-layout files.
+        # Deriving the separator from ``ai_skills`` would render such a
+        # ``.bob/commands/*.md`` (or ``.github/agents/*.agent.md``) file with
+        # ``/speckit-*`` whenever that agent is active in skills mode — even
+        # though a command-layout file must use ``/speckit.*``.  Deriving it
+        # from the agent's static output config avoids that mismatch and stays
+        # correct when a stale ``.bob/skills`` directory coexists with
+        # ``.bob/commands``.
+        _sep = agent_config.get("invoke_separator", ".")
+        try:
+            from specify_cli.integrations import get_integration  # noqa: PLC0415
+
+            _integ = get_integration(agent_name)
+            if _integ is not None:
+                registrar_writes_skills = (
+                    agent_config.get("extension") == "/SKILL.md"
+                )
+                _sep = _integ.invoke_separator_for_mode(registrar_writes_skills)
+        except Exception:
+            pass
+
         for cmd_info in commands:
             cmd_name = cmd_info["name"]
             aliases = cmd_info.get("aliases", [])
@@ -709,13 +743,18 @@ class CommandRegistrar:
             )
 
             # Resolve __SPECKIT_COMMAND_*__ tokens using the agent's invoke separator.
-            # The separator is sourced from agent_config (populated by _build_agent_configs,
-            # which propagates each integration's invoke_separator class attribute).
+            # For dual-layout agents (e.g. Bob) the separator differs between the
+            # skills and command layouts, so a single static AGENT_CONFIGS value is
+            # insufficient. ``_sep`` (resolved above) is derived from the *output
+            # layout* this registrar writes — a "/SKILL.md" scaffold uses the skills
+            # separator, any command-layout file uses the command separator — not
+            # the project's persisted ai_skills state. Single-layout agents fall back
+            # to the static AGENT_CONFIGS value unchanged (invoke_separator_for_mode
+            # default).
             # Deferred import of IntegrationBase avoids a circular import at module load
             # (base.py itself imports CommandRegistrar lazily).
             from specify_cli.integrations.base import IntegrationBase  # noqa: PLC0415
 
-            _sep = agent_config.get("invoke_separator", ".")
             body = IntegrationBase.resolve_command_refs(body, _sep)
 
             output_name = self._compute_output_name(agent_name, cmd_name, agent_config)

src/specify_cli/bundler/commands_impl/catalog_config.py

@@ -143,6 +143,13 @@ def add_source(
         raise BundlerError("A catalog url is required.")
     try:
         parsed = urlparse(url)
+        # Read .hostname inside the try: a bracketed-but-invalid IPv6 authority
+        # (e.g. "https://[not-an-ip]/c.json") parses cleanly under urlparse() on
+        # Python < 3.14 but raises ValueError lazily on the first .hostname access
+        # (the raise moved eager into urlparse() only in 3.14). Reading it here
+        # keeps that ValueError inside the guard instead of leaking a raw
+        # traceback past the CLI's `except BundlerError`. Reuse the value below.
+        hostname = parsed.hostname
     except ValueError as exc:
         raise BundlerError(f"Invalid catalog url: '{url}'.") from exc
     if not (parsed.scheme or parsed.path):
@@ -161,13 +168,13 @@ def add_source(
         # netloc — netloc is truthy for host-less URLs like "https://:8080"
         # or "https://user@". Validating here keeps junk out of
         # bundle-catalogs.yml instead of failing later at fetch time.
-        is_localhost = parsed.hostname in ("localhost", "127.0.0.1", "::1")
+        is_localhost = hostname in ("localhost", "127.0.0.1", "::1")
         if parsed.scheme.lower() != "https" and not is_localhost:
             raise BundlerError(
                 f"Catalog url must use HTTPS (got {parsed.scheme}://). "
                 "HTTP is only allowed for localhost."
             )
-        if not parsed.hostname:
+        if not hostname:
             raise BundlerError(f"Catalog url must be a valid URL with a host: {url}")
 
     url = _canonicalize_url(url)

src/specify_cli/bundler/models/catalog.py

@@ -43,7 +43,7 @@ class Scope(str, Enum):
 BUILTIN_DEFAULT_STACK: tuple[dict[str, Any], ...] = (
     {"id": "default", "url": "builtin://default", "priority": 1,
      "install_policy": InstallPolicy.INSTALL_ALLOWED.value},
-    {"id": "community", "url": "builtin://community", "priority": 2,
+    {"id": "community", "url": "builtin://community", "priority": 20,
      "install_policy": InstallPolicy.DISCOVERY_ONLY.value},
 )
 
@@ -251,8 +251,22 @@ def _merge_config(by_id: dict[str, CatalogSource], config_path: Path, scope: Sco
         return
     data = load_yaml(config_path)
     catalogs = data.get("catalogs") if isinstance(data, dict) else None
-    if not catalogs:
+    if catalogs is None:
         return
+    if not isinstance(catalogs, list):
+        # Treat only an absent/``None`` ``catalogs`` as "nothing to merge"; any
+        # other non-list value (``catalogs: 5``, ``false``, ``0``, ``''``,
+        # ``{}``) is a malformed config and must raise, not be silently skipped
+        # by a falsy check. Otherwise a truthy scalar would raise a raw
+        # ``TypeError: 'int' object is not iterable`` from the loop below, while
+        # falsy non-lists would be swallowed. Report the same actionable
+        # BundlerError the sibling reader of this file raises
+        # (commands_impl/catalog_config.py) so both readers of
+        # bundle-catalogs.yml agree. An empty list stays valid (loop is a no-op).
+        raise BundlerError(
+            f"Malformed catalog config at {config_path}: 'catalogs' must be a "
+            f"list, got {type(catalogs).__name__}."
+        )
     for raw in catalogs:
         src = CatalogSource.from_dict(raw, scope)
         by_id[src.id] = src

src/specify_cli/bundler/models/manifest.py

@@ -122,6 +122,11 @@ class BundleManifest:
 
         integration = None
         integration_raw = data.get("integration")
+        # Mirror the requires/provides guards above: a present-but-non-mapping
+        # 'integration' (e.g. a bare string "copilot") was silently dropped,
+        # leaving the bundle wrongly integration-agnostic. Reject it instead.
+        if integration_raw is not None and not isinstance(integration_raw, dict):
+            raise BundlerError("'integration' must be a mapping when present.")
         if isinstance(integration_raw, dict) and integration_raw.get("id"):
             integration = IntegrationRef(id=str(integration_raw["id"]).strip())
 

src/specify_cli/bundler/services/adapters.py

@@ -15,25 +15,26 @@ from pathlib import Path
 from urllib.parse import ParseResult, urlparse
 from urllib.request import url2pathname
 
+from ..._assets import _locate_core_pack, _repo_root
 from .. import BundlerError
 from ..lib.yamlio import loads_json
 from ..models.catalog import CatalogSource
 from ..models.manifest import ComponentRef
 
-# Built-in catalog payloads ship empty by default; a host distribution can
-# replace these with curated content. Keeping them here makes ``search``/``info``
-# work fully offline against the default stack.
+COMMUNITY_CATALOG_URL = (
+    "https://raw.githubusercontent.com/github/spec-kit/main/"
+    "bundles/catalog.community.json"
+)
+
+# The default catalog is reserved for first-party bundles. The community
+# catalog is loaded from the repository online and from the packaged snapshot
+# offline so discovery remains useful without network access.
 _BUILTIN_CATALOGS: dict[str, dict] = {
     "builtin://default": {
         "schema_version": "1.0",
         "catalog_url": "builtin://default",
         "bundles": {},
     },
-    "builtin://community": {
-        "schema_version": "1.0",
-        "catalog_url": "builtin://community",
-        "bundles": {},
-    },
 }
 
 HTTP_TIMEOUT_SECONDS = 10
@@ -95,6 +96,18 @@ def _validate_remote_url(source_id: str, url: str) -> None:
         )
 
 
+def _load_packaged_community_catalog() -> dict:
+    core_pack = _locate_core_pack()
+    path = (
+        core_pack / "bundles" / "catalog.community.json"
+        if core_pack is not None
+        else _repo_root() / "bundles" / "catalog.community.json"
+    )
+    if not path.is_file():
+        raise BundlerError(f"Bundled community catalog not found: {path}")
+    return loads_json(path.read_text(encoding="utf-8"), origin=str(path))
+
+
 def make_catalog_fetcher(*, allow_network: bool = True):
     """Return a fetcher callable suitable for :class:`CatalogStack`.
 
@@ -108,6 +121,10 @@ def make_catalog_fetcher(*, allow_network: bool = True):
         scheme = parsed.scheme.lower()
 
         if scheme == "builtin":
+            if url == "builtin://community":
+                if allow_network:
+                    return _http_get_json(source.id, COMMUNITY_CATALOG_URL)
+                return _load_packaged_community_catalog()
             payload = _BUILTIN_CATALOGS.get(url)
             if payload is None:
                 raise BundlerError(f"Unknown built-in catalog '{url}'.")

src/specify_cli/catalogs.py

@@ -149,7 +149,10 @@ class CatalogStackBase:
                 )
             try:
                 priority = int(raw_priority)
-            except (TypeError, ValueError):
+            except (TypeError, ValueError, OverflowError):
+                # OverflowError: int(float("inf")) — a YAML ``priority: .inf``
+                # would otherwise escape as an uncaught traceback instead of the
+                # clean validation error.
                 raise self._validation_error(
                     f"Invalid catalog config {config_path}: "
                     f"Invalid priority for catalog '{item.get('name', idx + 1)}': "

src/specify_cli/commands/init.py

@@ -86,7 +86,7 @@ def register(app: typer.Typer) -> None:
             help="Name for your new project directory (optional if using --here, or use '.' for current directory)",
         ),
         script_type: str = typer.Option(
-            None, "--script", help="Script type to use: sh or ps"
+            None, "--script", help="Script type to use: sh, ps, or py"
         ),
         ignore_agent_tools: bool = typer.Option(
             False,
@@ -458,6 +458,7 @@ def register(app: typer.Typer) -> None:
                     script_type=selected_script,
                     raw_options=integration_options,
                     parsed_options=integration_parsed_options or None,
+                    project_root=project_path,
                 )
                 _write_integration_json(
                     project_path,
@@ -478,7 +479,7 @@ def register(app: typer.Typer) -> None:
                     tracker=tracker,
                     force=force,
                     invoke_separator=resolved_integration.effective_invoke_separator(
-                        integration_parsed_options
+                        integration_parsed_options, project_root=project_path
                     ),
                 )
                 tracker.complete(
@@ -532,10 +533,8 @@ def register(app: typer.Typer) -> None:
                     "feature_numbering": "sequential",
                     "speckit_version": get_speckit_version(),
                 }
-                from ..integrations.base import SkillsIntegration as _SkillsPersist
-
-                if isinstance(resolved_integration, _SkillsPersist) or getattr(
-                    resolved_integration, "_skills_mode", False
+                if resolved_integration.is_skills_mode(
+                    integration_parsed_options or None, project_root=project_path
                 ):
                     init_opts["ai_skills"] = True
                 save_init_options(project_path, init_opts)
@@ -683,11 +682,9 @@ def register(app: typer.Typer) -> None:
             steps_lines.append("1. You're already in the project directory!")
             step_num = 2
 
-        from ..integrations.base import SkillsIntegration as _SkillsInt
-
-        _is_skills_integration = isinstance(
-            resolved_integration, _SkillsInt
-        ) or getattr(resolved_integration, "_skills_mode", False)
+        _is_skills_integration = resolved_integration.is_skills_mode(
+            integration_parsed_options or None, project_root=project_path
+        )
 
         codex_skill_mode = selected_ai == "codex" and _is_skills_integration
         zcode_skill_mode = selected_ai == "zcode" and _is_skills_integration
@@ -703,6 +700,8 @@ def register(app: typer.Typer) -> None:
         zed_skill_mode = selected_ai == "zed" and _is_skills_integration
         grok_skill_mode = selected_ai == "grok" and _is_skills_integration
         cline_skill_mode = selected_ai == "cline"
+        forge_skill_mode = selected_ai == "forge"
+        bob_skill_mode = selected_ai == "bob" and _is_skills_integration
         native_skill_mode = (
             codex_skill_mode
             or zcode_skill_mode
@@ -715,6 +714,7 @@ def register(app: typer.Typer) -> None:
             or devin_skill_mode
             or zed_skill_mode
             or grok_skill_mode
+            or bob_skill_mode
         )
 
         if codex_skill_mode:
@@ -752,6 +752,11 @@ def register(app: typer.Typer) -> None:
                 f"{step_num}. Start Grok Build in this project directory; spec-kit skills were installed to [cyan].grok/skills[/cyan]"
             )
             step_num += 1
+        if bob_skill_mode:
+            steps_lines.append(
+                f"{step_num}. Start Bob in this project directory; spec-kit skills were installed to [cyan].bob/skills[/cyan]"
+            )
+            step_num += 1
         usage_label = "skills" if native_skill_mode else "slash commands"
 
         from .._invocation_style import (
@@ -772,6 +777,7 @@ def register(app: typer.Typer) -> None:
             if (
                 _is_slash_skills_agent(selected_ai, _ai_skills_enabled)
                 or cline_skill_mode
+                or forge_skill_mode
             ):
                 return f"/speckit-{name}"
             return f"/speckit.{name}"

src/specify_cli/extensions/__init__.py

@@ -9,11 +9,13 @@ without bloating the core framework.
 from __future__ import annotations
 
 import copy
+import errno
 import hashlib
 import json
 import os
 import re
 import shutil
+import stat
 import tempfile
 import zipfile
 from dataclasses import dataclass
@@ -101,6 +103,51 @@ def _load_core_command_names() -> frozenset[str]:
 CORE_COMMAND_NAMES = _load_core_command_names()
 
 
+def _fsync_fd(fd: int) -> None:
+    """Sync a file descriptor, raising on real storage errors."""
+    try:
+        os.fsync(fd)
+    except AttributeError:
+        return
+    except NotImplementedError:
+        return
+    except OSError as exc:
+        if exc.errno in {errno.ENOTSUP, errno.EOPNOTSUPP, errno.EINVAL, errno.EBADF}:
+            return
+        raise
+
+
+def _fsync_directory(path: Path) -> None:
+    """Sync a directory when the platform supports it."""
+    if not path.exists():
+        return
+    if os.name == "nt":
+        return
+    try:
+        dir_fd = os.open(str(path), os.O_RDONLY | getattr(os, "O_DIRECTORY", 0))
+    except (AttributeError, NotImplementedError):
+        return
+    except OSError as exc:
+        if exc.errno in {errno.ENOTSUP, errno.EOPNOTSUPP, errno.EINVAL, errno.EBADF}:
+            return
+        try:
+            dir_fd = os.open(str(path), os.O_RDONLY)
+        except (AttributeError, NotImplementedError):
+            return
+        except OSError as exc2:
+            if exc2.errno in {errno.ENOTSUP, errno.EOPNOTSUPP, errno.EINVAL, errno.EBADF}:
+                return
+            raise
+    try:
+        _fsync_fd(dir_fd)
+    finally:
+        try:
+            os.close(dir_fd)
+        except OSError:
+            # Cleanup after an fsync failure should not mask the original error.
+            pass
+
+
 class ExtensionError(Exception):
     """Base exception for extension-related errors."""
 
@@ -136,7 +183,7 @@ def normalize_priority(value: Any, default: int = DEFAULT_HOOK_PRIORITY) -> int:
         return default
     try:
         priority = int(value)
-    except (TypeError, ValueError):
+    except (TypeError, ValueError, OverflowError):
         return default
     return priority if priority >= 1 else default
 
@@ -699,6 +746,55 @@ class ExtensionManager:
         self.extensions_dir = project_root / ".specify" / "extensions"
         self.registry = ExtensionRegistry(self.extensions_dir)
 
+    def _rescue_staging_dir(self, extension_id: str) -> Path:
+        """Fixed-length staging directory path for a preserved-config rescue.
+
+        The extension ID can be arbitrarily long (manifest validation caps only
+        the character set, not the length), so embedding it verbatim in a single
+        path component could push the ``.rescue-staging-<id>`` directory past a
+        filesystem's per-component byte limit and make every reinstall after
+        ``--keep-config`` fail with ``ENAMETOOLONG`` even though the extension
+        installs fine at ``dest_dir``. Hash the ID to a fixed-length suffix so
+        the component length is bounded regardless of ID length.
+        """
+        digest = hashlib.sha256(extension_id.encode("utf-8")).hexdigest()[:16]
+        return self.extensions_dir / f".rescue-staging-{digest}"
+
+    @staticmethod
+    def _has_keep_config_marker(directory: Path) -> bool:
+        """Return True when *directory* contains a valid ``.keep-config`` marker.
+
+        The marker is a regular (non-symlink) file written by
+        ``remove(..., keep_config=True)`` to record explicit provenance.  Its
+        content is intentionally empty — only presence matters, not content.
+        The symlink guard prevents a crafted symlink from fooling the check.
+        """
+        marker = directory / ".keep-config"
+        return marker.is_file() and not marker.is_symlink()
+
+    @staticmethod
+    def _is_legacy_keep_config_leftover(directory: Path) -> bool:
+        """Return True for the pre-marker ``remove(..., keep_config=True)`` layout.
+
+        Older CLI releases preserved only top-level config files and removed every
+        other entry, but they did not write ``.keep-config``. Recognize that exact
+        config-only leftover so upgrades still preserve user config, while
+        excluding partially-failed installs that still contain copied payload such
+        as ``extension.yml`` or command directories.
+        """
+        if not directory.is_dir() or directory.is_symlink():
+            return False
+
+        has_config = False
+        for entry in directory.iterdir():
+            if entry.name.endswith(("-config.yml", "-config.local.yml")) and (
+                entry.is_file() or entry.is_symlink()
+            ):
+                has_config = True
+                continue
+            return False
+        return has_config
+
     @staticmethod
     def _collect_manifest_command_names(manifest: ExtensionManifest) -> Dict[str, str]:
         """Collect command and alias names declared by a manifest.
@@ -1428,12 +1524,421 @@ class ExtensionManager:
                 backup_config_dir.unlink()
             did_remove = self.remove(manifest.id)
 
+        # Load and validate .extensionignore BEFORE reading/creating the rescue
+        # staging directory (and thus before deleting dest_dir). The loader can
+        # raise ValidationError (invalid UTF-8) or OSError; doing it first means
+        # such a failure aborts while the kept config is still authoritative in
+        # its documented location, rather than leaving a freshly published
+        # staging copy that a later retry (after the user edits the kept config)
+        # would reload and use to overwrite the newer bytes. Any staging left by
+        # an earlier destructive attempt is intentionally left intact here.
+        ignore_fn = self._load_extensionignore(source_dir)
+
+        # Rescue any config files left behind by a prior `remove --keep-config`.
+        # When an extension is removed with --keep-config, it is no longer in
+        # the registry but its config files remain in dest_dir.  A subsequent
+        # plain (non-force) install would delete that directory unconditionally,
+        # silently discarding the preserved config.  We read those files into
+        # memory and also write a durable staging copy outside dest_dir so
+        # that a partial rmtree, failed copytree, or partial restore cannot
+        # permanently discard the user's original bytes on a retry.  The
+        # staging dir is removed only after every config has been successfully
+        # restored.
+        stranded_configs: dict[str, tuple[bytes, int]] = {}
+        rescue_staging_dir = self._rescue_staging_dir(manifest.id)
+        # A staging directory is trusted only when this completion marker is
+        # present.  The marker is written after every staged file is complete
+        # and removed before the non-atomic cleanup, so a crash mid-staging or
+        # mid-cleanup can never leave a partial directory that a retry mistakes
+        # for a complete durable backup.
+        rescue_complete_marker = rescue_staging_dir / ".rescue-complete"
+        staging_is_complete = (
+            rescue_staging_dir.is_dir()
+            and not rescue_staging_dir.is_symlink()
+            and rescue_complete_marker.is_file()
+            and not rescue_complete_marker.is_symlink()
+        )
+
+        if staging_is_complete and not self.registry.is_installed(manifest.id):
+            # A previous install attempt staged the configs but never
+            # completed cleanly.  Reload from the durable backup so the
+            # original bytes are used on retry rather than whatever
+            # mixture of packaged defaults and partial restores remains
+            # on disk.  Only load non-symlinked files whose names match
+            # the two recognised config suffixes so a tampered staging
+            # directory cannot inject arbitrary files.
+            #
+            # A complete staging directory proves only that staging finished,
+            # not that dest_dir was ever modified: a crash after staging was
+            # synced but before the rmtree below leaves the live kept config
+            # intact. If the user then edits that live config before retrying,
+            # blindly preferring the staged bytes would silently overwrite the
+            # newer config. The staged and live copies are indistinguishable
+            # in provenance from disk alone (a genuine post-crash edit vs. a
+            # packaged default written by a partially-completed copytree), so
+            # when a live config disagrees with its staged copy we must not
+            # silently pick either — preserve both and abort, letting the user
+            # resolve it. dest_dir is still untouched here, so raising is safe.
+            def _recognized_config_names(
+                directory: Path, *, follow_symlinks: bool = True
+            ) -> set[str]:
+                names: set[str] = set()
+                if not directory.is_dir():
+                    return names
+                for entry in directory.iterdir():
+                    if not entry.name.endswith(
+                        ("-config.yml", "-config.local.yml")
+                    ):
+                        continue
+                    if follow_symlinks:
+                        if entry.is_file() and not entry.is_symlink():
+                            names.add(entry.name)
+                    else:
+                        # Include symlinks without following them so that
+                        # live-only symlinked configs are detected and
+                        # preserved rather than silently deleted.
+                        if entry.is_file() or entry.is_symlink():
+                            names.add(entry.name)
+                return names
+
+            conflicting: set[str] = set()
+            staged_names = _recognized_config_names(rescue_staging_dir)
+            live_names = _recognized_config_names(
+                dest_dir, follow_symlinks=False
+            )
+
+            def _matches_source_config_baseline(config_name: str) -> bool:
+                source_file = source_dir / config_name
+                live_file = dest_dir / config_name
+                if source_file.is_symlink() or live_file.is_symlink():
+                    return False
+                if not source_file.is_file() or not live_file.is_file():
+                    return False
+                try:
+                    source_stat = source_file.stat()
+                    source_bytes = source_file.read_bytes()
+                    live_stat = live_file.stat()
+                    live_bytes = live_file.read_bytes()
+                except OSError:
+                    return False
+                return live_bytes == source_bytes and stat.S_IMODE(
+                    live_stat.st_mode
+                ) == stat.S_IMODE(source_stat.st_mode)
+
+            # A live-only config created after the interrupted attempt is not
+            # enumerated by staging, so without this it would be silently
+            # deleted by the rmtree below and its bytes lost. Live-only files
+            # that still match the current package baseline are safe: they were
+            # copied by the interrupted install and can be recreated on retry.
+            # Only truly divergent live-only configs are conflicts.
+            live_only = live_names - staged_names
+            conflicting.update(
+                name
+                for name in live_only
+                if not _matches_source_config_baseline(name)
+            )
+            # Load original permission bits from the sidecar JSON written by
+            # the staging step.  Staged files are kept at mode 0o600 so that
+            # rmtree always succeeds on Windows, so staged_stat.st_mode would
+            # always be 0o600 and must not be used for mode comparisons or
+            # restoration; the sidecar records the true original mode.
+            rescue_modes_file = rescue_staging_dir / ".rescue-modes.json"
+            _staged_modes: dict[str, int] = {}
+            if rescue_modes_file.is_file() and not rescue_modes_file.is_symlink():
+                try:
+                    _loaded_modes = json.loads(rescue_modes_file.read_bytes())
+                except (OSError, ValueError):
+                    # Ignore unreadable/invalid sidecar metadata and fall back
+                    # to each staged file's mode for compatibility.
+                    pass
+                else:
+                    # json.loads() succeeds for any valid JSON document, so a
+                    # sidecar containing e.g. `[]` or a string would otherwise
+                    # crash later at _staged_modes.get() or stat.S_IMODE().
+                    # Accept only a mapping of string filenames to integer modes
+                    # (bool is rejected despite subclassing int); anything else
+                    # falls back to each staged file's own mode.
+                    if isinstance(_loaded_modes, dict) and all(
+                        isinstance(name, str)
+                        and isinstance(recorded_mode, int)
+                        and not isinstance(recorded_mode, bool)
+                        for name, recorded_mode in _loaded_modes.items()
+                    ):
+                        _staged_modes = _loaded_modes
+            for staged_name in sorted(staged_names):
+                staged_file = rescue_staging_dir / staged_name
+                staged_stat = staged_file.stat()
+                staged_bytes = staged_file.read_bytes()
+                # Prefer the sidecar-recorded mode; fall back to the staged
+                # file's own mode for backwards-compat with staging dirs
+                # written before the sidecar was introduced.
+                staged_mode = _staged_modes.get(
+                    staged_name, stat.S_IMODE(staged_stat.st_mode)
+                )
+                live_file = dest_dir / staged_name
+                if live_file.is_symlink():
+                    # A user may have replaced the live config with a symlink
+                    # after the interrupted attempt. It cannot be compared by
+                    # bytes/mode against the staged copy, and the rmtree below
+                    # would silently delete this newer choice and restore the
+                    # older staged file. Treat any live symlink as a conflict so
+                    # both are preserved and the user resolves it.
+                    conflicting.add(staged_name)
+                elif live_file.is_file():
+                    # A live config that cannot be read or stat'ed must not be
+                    # treated as non-conflicting: the rmtree below would delete
+                    # it and restore the stale staged copy. Abort while dest_dir
+                    # is untouched so no newer or permission-restricted config is
+                    # lost. Divergence also includes permission-only edits (for
+                    # example tightening a secret-bearing config from 0644 to
+                    # 0600), which byte equality alone would miss and then revert.
+                    try:
+                        live_stat = live_file.stat()
+                        live_bytes = live_file.read_bytes()
+                    except OSError:
+                        conflicting.add(staged_name)
+                    else:
+                        if live_bytes != staged_bytes or stat.S_IMODE(
+                            live_stat.st_mode
+                        ) != staged_mode:
+                            conflicting.add(staged_name)
+                stranded_configs[staged_name] = (staged_bytes, staged_mode)
+            if conflicting:
+                # Split into two cases for accurate user guidance: files that
+                # exist in both locations but have diverged, and files that
+                # exist only in the live directory with no rescue-backup copy.
+                both_diverged = conflicting - live_only
+                live_only_conflict = conflicting & live_only
+                msg_parts: list[str] = [
+                    f"Preserved extension config conflict for '{manifest.id}':"
+                ]
+                if both_diverged:
+                    names = ", ".join(sorted(both_diverged))
+                    msg_parts.append(
+                        f"The current config(s) ({names}) in {dest_dir} differ"
+                        f" from their rescued backup in {rescue_staging_dir}."
+                        " Both copies have been preserved."
+                    )
+                if live_only_conflict:
+                    names = ", ".join(sorted(live_only_conflict))
+                    msg_parts.append(
+                        f"The config(s) ({names}) exist only in {dest_dir}"
+                        f" with no counterpart in the rescued backup at"
+                        f" {rescue_staging_dir}."
+                    )
+                msg_parts.append(
+                    f"Reconcile {dest_dir} and {rescue_staging_dir} to the"
+                    f" desired final state, delete {rescue_staging_dir},"
+                    " then reinstall."
+                )
+                raise ValidationError(" ".join(msg_parts))
+        elif (
+            dest_dir.exists()
+            and not self.registry.is_installed(manifest.id)
+            and (
+                self._has_keep_config_marker(dest_dir)
+                or self._is_legacy_keep_config_leftover(dest_dir)
+            )
+        ):
+            for cfg_file in (
+                list(dest_dir.glob("*-config.yml"))
+                + list(dest_dir.glob("*-config.local.yml"))
+            ):
+                if cfg_file.is_symlink():
+                    # `remove --keep-config` preserves a symlinked config
+                    # because Path.is_file() follows symlinks. Its bytes cannot
+                    # be safely rescued (the target may live outside dest_dir),
+                    # and the rmtree below would delete the link and silently
+                    # discard the kept configuration. Reject the reinstall while
+                    # dest_dir is untouched so the user resolves it rather than
+                    # losing the linked config.
+                    raise ValidationError(
+                        "Preserved extension config for "
+                        f"'{manifest.id}' is a symlink ({cfg_file.name}) in "
+                        f"{dest_dir}, which cannot be safely rescued during "
+                        "reinstall. Resolve manually — replace the symlink with "
+                        "a regular file or remove it — then reinstall."
+                    )
+                if cfg_file.is_file():
+                    stranded_configs[cfg_file.name] = (
+                        cfg_file.read_bytes(),
+                        cfg_file.stat().st_mode,
+                    )
+
+        if stranded_configs and not staging_is_complete:
+            # Write a durable backup outside dest_dir before any
+            # destructive operation so the original bytes survive a
+            # crash or partial failure at any later step.  The staging
+            # dir is cleaned up only after every restore succeeds.
+            #
+            # Any pre-existing staging dir here lacks the completion marker
+            # (staging_is_complete is False), so it is a stale partial from an
+            # interrupted attempt — remove it first for a clean write.
+            if rescue_staging_dir.is_symlink():
+                rescue_staging_dir.unlink()
+            elif rescue_staging_dir.is_dir():
+                shutil.rmtree(rescue_staging_dir)
+            elif rescue_staging_dir.exists():
+                rescue_staging_dir.unlink()
+            try:
+                rescue_staging_dir.mkdir(parents=True, exist_ok=True)
+                for filename, (content, mode) in stranded_configs.items():
+                    staged = rescue_staging_dir / filename
+                    # Create the staging file with mode 0600 before writing so
+                    # the preserved bytes are never transiently readable by other
+                    # local users, even on a umask that would produce 0644.
+                    # O_BINARY (0 on POSIX) is required so Windows does not open
+                    # the descriptor in text mode and translate the preserved
+                    # bytes' "\n" into "\r\n" as they are written.
+                    fd = os.open(
+                        str(staged),
+                        os.O_WRONLY | os.O_CREAT | os.O_EXCL | getattr(os, "O_BINARY", 0),
+                        0o600,
+                    )
+                    try:
+                        # os.write() may write fewer bytes than requested, so
+                        # loop until the whole buffer is on disk — a truncated
+                        # "durable" backup would be trusted over the intact
+                        # config on a retry and cause silent data loss.
+                        view = memoryview(content)
+                        written = 0
+                        while written < len(view):
+                            written += os.write(fd, view[written:])
+                        # Do NOT chmod the staged file: setting a read-only
+                        # mode (e.g. 0o444) makes the file undeletable on
+                        # Windows and causes shutil.rmtree to fail during
+                        # cleanup.  Original modes are recorded separately in
+                        # .rescue-modes.json so they can be reapplied when the
+                        # config is actually restored.
+                        _fsync_fd(fd)
+                    finally:
+                        os.close(fd)
+                # Persist the original permission bits in a sidecar JSON file
+                # so a retry can correctly reapply them even though the staged
+                # files themselves are kept at their creation mode (0o600).
+                rescue_modes_file = rescue_staging_dir / ".rescue-modes.json"
+                modes_payload = json.dumps(
+                    {
+                        filename: stat.S_IMODE(mode)
+                        for filename, (_, mode) in stranded_configs.items()
+                    },
+                    sort_keys=True,
+                ).encode()
+                modes_fd = os.open(
+                    str(rescue_modes_file),
+                    os.O_WRONLY | os.O_CREAT | os.O_EXCL | getattr(os, "O_BINARY", 0),
+                    0o600,
+                )
+                try:
+                    view = memoryview(modes_payload)
+                    written = 0
+                    while written < len(view):
+                        written += os.write(modes_fd, view[written:])
+                    _fsync_fd(modes_fd)
+                finally:
+                    os.close(modes_fd)
+                # Flush the staging directory metadata before publishing the
+                # completion marker so a crash cannot leave a visible marker with
+                # only a subset of staged files.
+                _fsync_directory(rescue_staging_dir)
+                # Write the completion marker only after every staged file is
+                # fully written so a retry trusts staging only when it is whole.
+                marker_fd = os.open(
+                    str(rescue_complete_marker),
+                    os.O_WRONLY | os.O_CREAT | os.O_EXCL,
+                    0o600,
+                )
+                try:
+                    _fsync_fd(marker_fd)
+                finally:
+                    os.close(marker_fd)
+                _fsync_directory(rescue_staging_dir)
+                _fsync_directory(rescue_staging_dir.parent)
+            except BaseException:
+                # Durable staging failed (or was interrupted).  Continuing with
+                # only the in-memory copy would reintroduce the permanent-loss
+                # path this staging exists to close: the rmtree below could
+                # delete the originals and a later restore failure would leave
+                # no on-disk copy.  dest_dir is still untouched here, so clean
+                # up the partial staging dir and abort the install instead of
+                # proceeding destructively.
+                shutil.rmtree(rescue_staging_dir, ignore_errors=True)
+                raise
+
         # Install extension (dest_dir computed above during self-install guard)
         if dest_dir.exists():
             shutil.rmtree(dest_dir)
 
-        ignore_fn = self._load_extensionignore(source_dir)
-        shutil.copytree(source_dir, dest_dir, ignore=ignore_fn)
+        def _restore_stranded_config_file(
+            target: Path, content: bytes, preserved_mode: int
+        ) -> None:
+            tmp_path: Path | None = None
+            try:
+                # A short fixed prefix, not f".{target.name}.": the preserved
+                # config filename may itself already be near the filesystem's
+                # per-component byte limit, and NamedTemporaryFile appends a
+                # random suffix to the prefix — reusing the full name would push
+                # the temp file past the limit and raise ENAMETOOLONG on every
+                # retry. tempfile already guarantees collision avoidance.
+                with tempfile.NamedTemporaryFile(
+                    mode="wb",
+                    dir=target.parent,
+                    prefix=".cfg-restore.",
+                    delete=False,
+                ) as tmp:
+                    tmp_path = Path(tmp.name)
+                    tmp.write(content)
+                    tmp.flush()
+                    _fsync_fd(tmp.fileno())
+                try:
+                    tmp_path.chmod(stat.S_IMODE(preserved_mode))
+                except (NotImplementedError, OSError):
+                    pass  # Best-effort; chmod may not be supported on all platforms.
+                os.replace(tmp_path, target)
+                try:
+                    target_fd = os.open(str(target), os.O_RDONLY)
+                except (AttributeError, OSError, NotImplementedError):
+                    target_fd = None
+                try:
+                    if target_fd is not None:
+                        _fsync_fd(target_fd)
+                finally:
+                    if target_fd is not None:
+                        try:
+                            os.close(target_fd)
+                        except OSError:
+                            pass  # best-effort close during cleanup; ignore errors
+                _fsync_directory(target.parent)
+            except BaseException:
+                if tmp_path is not None and tmp_path.exists():
+                    tmp_path.unlink()
+                raise
+
+        try:
+            shutil.copytree(source_dir, dest_dir, ignore=ignore_fn)
+        except BaseException:
+            # copytree failed — dest_dir may be absent or only partially
+            # created.  Write the rescued configs back now so they are not
+            # permanently lost even though the install did not complete.
+            if stranded_configs:
+                dest_dir.mkdir(parents=True, exist_ok=True)
+                for filename, (content, mode) in stranded_configs.items():
+                    target = dest_dir / filename
+                    _restore_stranded_config_file(target, content, mode)
+            raise
+
+        # Restore stranded configs rescued before the rmtree above.
+        for filename, (content, mode) in stranded_configs.items():
+            target = dest_dir / filename
+            _restore_stranded_config_file(target, content, mode)
+
+        # NOTE: the durable staging backup is intentionally NOT cleaned up
+        # here.  Command/skill/hook registration and the final registry.add()
+        # below can still fail; if we discarded the backup and provenance now,
+        # such a failure would leave the extension unregistered with no durable
+        # rescue copy, so the next plain retry would skip rescue and overwrite
+        # the restored user config with packaged defaults.  Cleanup is deferred
+        # until after registry.add() succeeds (see post-commit cleanup below).
 
         # Register commands with AI agents
         registered_commands = {}
@@ -1496,6 +2001,24 @@ class ExtensionManager:
             },
         )
 
+        # Post-commit cleanup: the registry now records this extension as
+        # installed, so the rescue guard (`not self.registry.is_installed`)
+        # will never misread a leftover staging dir on a future run.  The
+        # durable backup has therefore served its purpose and can be removed
+        # best-effort — a cleanup failure must not fail an install that has
+        # already committed successfully.
+        if rescue_staging_dir.is_dir() and not rescue_staging_dir.is_symlink():
+            # Remove the completion marker before the non-atomic rmtree so a
+            # crash mid-cleanup cannot leave a staging dir that a retry would
+            # wrongly trust as a complete durable backup.
+            try:
+                rescue_complete_marker.unlink(missing_ok=True)
+                _fsync_directory(rescue_staging_dir)
+                shutil.rmtree(rescue_staging_dir)
+                _fsync_directory(rescue_staging_dir.parent)
+            except OSError:
+                pass  # Best-effort; install already committed to the registry.
+
         return manifest
 
     def install_from_zip(
@@ -1612,6 +2135,12 @@ class ExtensionManager:
                         shutil.rmtree(child)
                     else:
                         child.unlink()
+                # Write a provenance marker so install_from_directory can
+                # distinguish this --keep-config leftover from a directory left
+                # by a partially-failed install (which must not have its
+                # packaged default configs treated as user-preserved data).
+                # Content is intentionally empty — only presence matters.
+                (extension_dir / ".keep-config").write_text("")
         else:
             # Backup config files before deleting
             if extension_dir.exists():
@@ -2073,14 +2602,23 @@ class ExtensionCatalog(CatalogStackBase):
         url: str,
         timeout: int = 10,
         extra_headers: Optional[Dict[str, str]] = None,
+        redirect_validator=None,
     ):
         """Open a URL with provider-based auth, trying each configured provider.
 
         Delegates to :func:`specify_cli.authentication.http.open_url`.
+        *redirect_validator*, when provided, is invoked as ``(old_url, new_url)``
+        before EACH redirect hop so an HTTPS host guarantee can be enforced on
+        every intermediate URL, not just the terminal one.
         """
         from specify_cli.authentication.http import open_url
 
-        return open_url(url, timeout, extra_headers=extra_headers)
+        return open_url(
+            url,
+            timeout,
+            extra_headers=extra_headers,
+            redirect_validator=redirect_validator,
+        )
 
     def _resolve_github_release_asset_api_url(
         self,
@@ -2304,7 +2842,24 @@ class ExtensionCatalog(CatalogStackBase):
 
         # Fetch from network
         try:
-            with self._open_url(entry.url, timeout=10) as response:
+            # Validate EVERY redirect hop, not just the terminal URL. _open_url
+            # follows redirects; _StripAuthOnRedirect drops auth on an HTTPS->HTTP
+            # downgrade AND whenever the redirect leaves the configured trusted
+            # hosts, but the payload itself is still fetched and trusted, and it
+            # supplies each extension's download_url + sha256 (so a redirected
+            # payload defeats sha256 verification). A terminal-only check also
+            # misses an https -> http -> attacker-https chain. redirect_validator
+            # runs before each hop; the final geturl() check is kept as a
+            # belt-and-braces guard. Mirrors bundler/services/adapters.py.
+            def _validate_redirect(_old_url: str, new_url: str) -> None:
+                self._validate_catalog_url(new_url)
+
+            with self._open_url(
+                entry.url, timeout=10, redirect_validator=_validate_redirect
+            ) as response:
+                final_url = response.geturl()
+                if final_url != entry.url:
+                    self._validate_catalog_url(final_url)
                 catalog_data = json.loads(response.read())
 
             self._validate_catalog_payload(catalog_data, entry.url)
@@ -2481,7 +3036,18 @@ class ExtensionCatalog(CatalogStackBase):
         try:
             import urllib.error
 
-            with self._open_url(catalog_url, timeout=10) as response:
+            # Same redirect hardening as _fetch_single_catalog: validate every
+            # redirect hop AND the final URL so this legacy single-catalog path
+            # is not vulnerable to an HTTPS->HTTP redirected payload either.
+            def _validate_redirect(_old_url: str, new_url: str) -> None:
+                self._validate_catalog_url(new_url)
+
+            with self._open_url(
+                catalog_url, timeout=10, redirect_validator=_validate_redirect
+            ) as response:
+                final_url = response.geturl()
+                if final_url != catalog_url:
+                    self._validate_catalog_url(final_url)
                 catalog_data = json.loads(response.read())
 
             # Validate catalog structure. Reuses the same helper as
@@ -2631,8 +3197,20 @@ class ExtensionCatalog(CatalogStackBase):
         # Validate download URL requires HTTPS (prevent man-in-the-middle attacks)
         from urllib.parse import urlparse
 
-        parsed = urlparse(download_url)
-        is_localhost = parsed.hostname in ("localhost", "127.0.0.1", "::1")
+        # A malformed authority (e.g. an unterminated IPv6 bracket
+        # "https://[::1") makes urlparse / hostname access raise ValueError.
+        # The download_url comes from catalog payload data, so surface a clean
+        # ExtensionError rather than leaking a raw ValueError past the command
+        # handler (which only catches ExtensionError). Mirrors catalogs (#3435)
+        # and workflows/catalog.py (#3484).
+        try:
+            parsed = urlparse(download_url)
+            hostname = parsed.hostname
+        except ValueError:
+            raise ExtensionError(
+                f"Extension download URL is malformed: {download_url}"
+            ) from None
+        is_localhost = hostname in ("localhost", "127.0.0.1", "::1")
         if parsed.scheme != "https" and not (parsed.scheme == "http" and is_localhost):
             raise ExtensionError(
                 f"Extension download URL must use HTTPS: {download_url}"
@@ -3030,6 +3608,7 @@ class HookExecutor:
         dollar_skill_mode = is_dollar_skills_agent(selected_ai, ai_skills_enabled)
         kimi_skill_mode = selected_ai == "kimi"
         cline_mode = selected_ai == "cline"
+        forge_mode = selected_ai == "forge"
 
         skill_name = self._skill_name_from_command(command_id)
         if dollar_skill_mode and skill_name:
@@ -3040,6 +3619,10 @@ class HookExecutor:
             from ..integrations.cline import format_cline_command_name
 
             return f"/{format_cline_command_name(command_id)}"
+        if forge_mode:
+            from ..integrations.forge import format_forge_command_name
+
+            return f"/{format_forge_command_name(command_id)}"
 
         use_slash = is_slash_skills_agent(selected_ai, ai_skills_enabled)
 

src/specify_cli/integration_runtime.py

@@ -46,6 +46,7 @@ def with_integration_setting(
     script_type: str | None = None,
     raw_options: str | None = None,
     parsed_options: dict[str, Any] | None = None,
+    project_root: Any = None,
 ) -> dict[str, dict[str, Any]]:
     """Return integration settings with *key* updated."""
     settings = integration_settings(state)
@@ -63,7 +64,9 @@ def with_integration_setting(
     elif raw_options is not None:
         current.pop("parsed_options", None)
 
-    current["invoke_separator"] = integration.effective_invoke_separator(parsed_options)
+    current["invoke_separator"] = integration.effective_invoke_separator(
+        parsed_options, project_root
+    )
     settings[key] = current
     return settings
 
@@ -73,10 +76,11 @@ def invoke_separator_for_integration(
     state: dict[str, Any],
     key: str,
     parsed_options: dict[str, Any] | None = None,
+    project_root: Any = None,
 ) -> str:
     """Resolve the invocation separator for stored/default integration state."""
     if parsed_options is not None:
-        return integration.effective_invoke_separator(parsed_options)
+        return integration.effective_invoke_separator(parsed_options, project_root)
 
     setting = integration_setting(state, key)
     stored_separator = setting.get("invoke_separator")
@@ -85,6 +89,6 @@ def invoke_separator_for_integration(
 
     stored_parsed = setting.get("parsed_options")
     if isinstance(stored_parsed, dict):
-        return integration.effective_invoke_separator(stored_parsed)
+        return integration.effective_invoke_separator(stored_parsed, project_root)
 
-    return integration.effective_invoke_separator(None)
+    return integration.effective_invoke_separator(None, project_root)

src/specify_cli/integrations/__init__.py

@@ -58,6 +58,7 @@ def _register_builtins() -> None:
     from .copilot import CopilotIntegration
     from .cursor_agent import CursorAgentIntegration
     from .devin import DevinIntegration
+    from .droid import DroidIntegration
     from .firebender import FirebenderIntegration
     from .forge import ForgeIntegration
     from .gemini import GeminiIntegration
@@ -95,6 +96,7 @@ def _register_builtins() -> None:
     _register(CopilotIntegration())
     _register(CursorAgentIntegration())
     _register(DevinIntegration())
+    _register(DroidIntegration())
     _register(FirebenderIntegration())
     _register(ForgeIntegration())
     _register(GeminiIntegration())

src/specify_cli/integrations/_helpers.py

@@ -6,6 +6,7 @@ from pathlib import Path
 from typing import Any, Callable
 
 import typer
+from rich.markup import escape
 
 from .._agent_config import SCRIPT_TYPE_CHOICES
 from .._console import console
@@ -206,7 +207,7 @@ def _parse_integration_options(integration: Any, raw_options: str) -> dict[str,
     while i < len(tokens):
         token = tokens[i]
         if not token.startswith("-"):
-            console.print(f"[red]Error:[/red] Unexpected integration option value '{token}'.")
+            console.print(f"[red]Error:[/red] Unexpected integration option value '{escape(token)}'.")
             if allowed:
                 console.print(f"Allowed options: {allowed}")
             raise typer.Exit(1)
@@ -217,7 +218,7 @@ def _parse_integration_options(integration: Any, raw_options: str) -> dict[str,
             name, value = name.split("=", 1)
         opt = declared.get(name)
         if not opt:
-            console.print(f"[red]Error:[/red] Unknown integration option '{token}'.")
+            console.print(f"[red]Error:[/red] Unknown integration option '{escape(token)}'.")
             if allowed:
                 console.print(f"Allowed options: {allowed}")
             raise typer.Exit(1)
@@ -272,24 +273,19 @@ def _update_init_options_for_integration(
         load_init_options,
         save_init_options,
     )
-    from .base import SkillsIntegration
     opts = load_init_options(project_root)
     opts["integration"] = integration.key
     opts["ai"] = integration.key
     opts["speckit_version"] = _get_speckit_version()
     if script_type:
         opts["script"] = script_type
-    # Skills mode is either intrinsic (SkillsIntegration), set on the instance
-    # during setup() (_skills_mode), or requested via parsed options (e.g.
-    # Copilot's --skills, persisted as parsed_options["skills"]). The latter is
-    # the only signal available on the `use` path, where no setup() runs and a
-    # fresh integration instance has _skills_mode == False (issue #3550).
-    skills_mode = (
-        isinstance(integration, SkillsIntegration)
-        or getattr(integration, "_skills_mode", False)
-        or bool((parsed_options or {}).get("skills"))
-    )
-    if skills_mode:
+    # Whether skills mode is active is owned by each integration via the
+    # ``is_skills_mode`` hook (base default honors ``--skills``;
+    # SkillsIntegration returns True; skills-first integrations with a legacy
+    # opt-out such as Bob override it). This keeps shared code free of
+    # ``isinstance`` / ``_skills_mode`` probing. Passing parsed_options lets it
+    # work on the ``use``/``install`` path where no setup() runs (issue #3550).
+    if integration.is_skills_mode(parsed_options, project_root=project_root):
         opts["ai_skills"] = True
     else:
         opts.pop("ai_skills", None)
@@ -325,6 +321,7 @@ def _set_default_integration(
         script_type=resolved_script,
         raw_options=raw_options,
         parsed_options=parsed_options,
+        project_root=project_root,
     )
 
     if refresh_templates:
@@ -333,7 +330,8 @@ def _set_default_integration(
                 project_root,
                 resolved_script,
                 invoke_separator=_invoke_separator_for_integration(
-                    integration, {"integration_settings": settings}, key, parsed_options
+                    integration, {"integration_settings": settings}, key, parsed_options,
+                    project_root=project_root,
                 ),
                 force=refresh_templates_force,
                 refresh_managed=True,

src/specify_cli/integrations/_install_commands.py

@@ -38,7 +38,7 @@ from ._helpers import (
 @integration_app.command("install")
 def integration_install(
     key: str = typer.Argument(help="Integration key to install (e.g. claude, copilot)"),
-    script: str | None = typer.Option(None, "--script", help="Script type: sh or ps (default: from init-options.json or platform default)"),
+    script: str | None = typer.Option(None, "--script", help="Script type: sh, ps, or py (default: from init-options.json or platform default)"),
     force: bool = typer.Option(False, "--force", help="Allow multi-install when integrations are not declared safe"),
     integration_options: str | None = typer.Option(None, "--integration-options", help='Options for the integration (e.g. --integration-options="--commands-dir .myagent/cmds")'),
 ):
@@ -127,7 +127,8 @@ def integration_install(
         project_root,
         selected_script,
         invoke_separator=_invoke_separator_for_integration(
-            infra_integration, current, infra_key, infra_parsed
+            infra_integration, current, infra_key, infra_parsed,
+            project_root=project_root,
         ),
     )
     if os.name != "nt":
@@ -155,10 +156,16 @@ def integration_install(
             script_type=selected_script,
             raw_options=raw_options,
             parsed_options=parsed_options,
+            project_root=project_root,
         )
         _write_integration_json(project_root, new_default, new_installed, settings)
         if new_default == integration.key:
-            _update_init_options_for_integration(project_root, integration, script_type=selected_script)
+            _update_init_options_for_integration(
+                project_root,
+                integration,
+                script_type=selected_script,
+                parsed_options=parsed_options,
+            )
         else:
             _refresh_init_options_speckit_version(project_root)
 

src/specify_cli/integrations/_migrate_commands.py

@@ -1,8 +1,9 @@
 """specify integration switch / upgrade command handlers."""
 from __future__ import annotations
 
+import json
 import os
-from pathlib import PurePath
+from pathlib import Path, PurePath
 
 import typer
 
@@ -40,10 +41,97 @@ from ._helpers import (
 )
 
 
+def _manifest_tracks_skill_layout(manifest) -> bool:
+    """Return True when *manifest* tracks any skills-layout artifact.
+
+    A skill scaffold is written as ``.../speckit-<name>/SKILL.md``, so a
+    manifest whose tracked files include a ``/SKILL.md`` key is in the skills
+    layout; otherwise it is in the command layout. Used by ``upgrade`` to
+    detect a dual-mode agent (e.g. Bob) flipping between the legacy commands
+    layout and the skills layout so orphaned extension artifacts from the old
+    layout can be reconciled.
+    """
+    return any(str(rel).endswith("/SKILL.md") for rel in manifest.files)
+
+
+class _PresetRegistryUnreadableError(Exception):
+    """Raised when an existing preset registry cannot be read or parsed.
+
+    Distinct from a *genuinely absent* registry (no presets installed): an
+    unreadable registry means we cannot verify whether preset overrides would
+    be orphaned by a layout change, so the migration must be rejected rather
+    than proceeding on a false "no presets" assumption.
+    """
+
+
+def _installed_presets_affecting_agent(project_root, agent_key: str) -> list[str]:
+    """Return IDs of installed presets with artifacts registered for *agent_key*.
+
+    Presets register command overrides for every detected agent and mirror
+    skills for the active skills agent, tracking the result in each preset's
+    ``registered_commands`` / ``registered_skills`` metadata. There is no
+    agent-scoped preset re-registration mechanism, so a command↔skills *layout
+    change* cannot reconcile those artifacts (see ``integration_upgrade``).
+    Callers use this to detect the unsafe case and reject the migration rather
+    than silently orphaning preset files / leaving stale registry entries.
+
+    Fails **closed**: a genuinely absent registry (no presets ever installed)
+    returns an empty list, but if the registry file exists and cannot be read
+    or parsed (e.g. a permission error or corruption) this raises
+    :class:`_PresetRegistryUnreadableError`.  Reporting "no presets" in that
+    case would let a ``--force`` layout-changing upgrade delete
+    preset-overridden files while their registry state can't be reconciled —
+    the exact inconsistency the guard exists to prevent.
+    """
+    from ..presets import PresetRegistry
+
+    registry_path = (
+        Path(project_root) / ".specify" / "presets" / PresetRegistry.REGISTRY_FILE
+    )
+    # Genuinely absent registry → no presets installed → safe to proceed.
+    if not registry_path.exists():
+        return []
+
+    # The registry exists: any failure to read or parse it must surface as an
+    # error, not be swallowed into an empty ("no presets") result.
+    try:
+        data = json.loads(registry_path.read_text(encoding="utf-8"))
+    except (OSError, ValueError) as exc:
+        raise _PresetRegistryUnreadableError(str(exc)) from exc
+    if not isinstance(data, dict) or not isinstance(data.get("presets", {}), dict):
+        raise _PresetRegistryUnreadableError(
+            "preset registry structure is malformed"
+        )
+
+    affected: list[str] = []
+    for preset_id, meta in data.get("presets", {}).items():
+        # A malformed entry means we cannot verify whether this preset owns
+        # artifacts for the agent, so fail closed rather than skip it.
+        if not isinstance(meta, dict):
+            raise _PresetRegistryUnreadableError(
+                f"preset '{preset_id}' entry is malformed"
+            )
+        registered_commands = meta.get("registered_commands", {})
+        if not isinstance(registered_commands, dict):
+            raise _PresetRegistryUnreadableError(
+                f"preset '{preset_id}' registered_commands is malformed"
+            )
+        registered_skills = meta.get("registered_skills", [])
+        if not isinstance(registered_skills, (list, tuple)):
+            raise _PresetRegistryUnreadableError(
+                f"preset '{preset_id}' registered_skills is malformed"
+            )
+        has_commands = bool(registered_commands.get(agent_key))
+        has_skills = bool(registered_skills)
+        if has_commands or has_skills:
+            affected.append(preset_id)
+    return affected
+
+
 @integration_app.command("switch")
 def integration_switch(
     target: str = typer.Argument(help="Integration key to switch to"),
-    script: str | None = typer.Option(None, "--script", help="Script type: sh or ps (default: from init-options.json or platform default)"),
+    script: str | None = typer.Option(None, "--script", help="Script type: sh, ps, or py (default: from init-options.json or platform default)"),
     force: bool = typer.Option(False, "--force", help="Force removal of modified files during uninstall of the previous integration"),
     refresh_shared_infra: bool = typer.Option(False, "--refresh-shared-infra", help="Also overwrite shared infrastructure files even if you customized them (otherwise customizations are preserved)"),
     integration_options: str | None = typer.Option(None, "--integration-options", help='Options for the target integration'),
@@ -236,7 +324,8 @@ def integration_switch(
         force=refresh_shared_infra,
         refresh_managed=True,
         invoke_separator=_invoke_separator_for_integration(
-            target_integration, current, target, parsed_options
+            target_integration, current, target, parsed_options,
+            project_root=project_root,
         ),
         refresh_hint=(
             "To overwrite customizations, re-run with "
@@ -336,7 +425,7 @@ def integration_switch(
 def integration_upgrade(
     key: str | None = typer.Argument(None, help="Integration key to upgrade (default: current integration)"),
     force: bool = typer.Option(False, "--force", help="Force upgrade even if files are modified"),
-    script: str | None = typer.Option(None, "--script", help="Script type: sh or ps (default: from init-options.json or platform default)"),
+    script: str | None = typer.Option(None, "--script", help="Script type: sh, ps, or py (default: from init-options.json or platform default)"),
     integration_options: str | None = typer.Option(None, "--integration-options", help="Options for the integration"),
 ):
     """Upgrade an integration by reinstalling with diff-aware file handling.
@@ -398,6 +487,56 @@ def integration_upgrade(
         integration, current, key, integration_options
     )
 
+    # Guard: reject a command↔skills layout change while preset overrides are
+    # installed for this agent (review #3415).  A dual-mode agent (e.g. Bob)
+    # can flip layout across an upgrade (``--skills`` / ``--legacy-commands``).
+    # Extension artifacts are reconciled after the flip (see below), but preset
+    # artifacts cannot be: there is no agent-scoped preset re-registration
+    # anywhere in the CLI, so migrating would delete a preset's old-layout
+    # files without recreating them in the new layout and leave the preset
+    # registry claiming artifacts that no longer exist.  Detect the intended
+    # layout (``is_skills_mode`` reflects the resolved flags/disk state, so a
+    # plain same-layout upgrade is unaffected) and bail out *before* any
+    # mutation with an actionable error so the project is never left in a
+    # half-migrated, inconsistent state.
+    if _manifest_tracks_skill_layout(old_manifest) != integration.is_skills_mode(
+        parsed_options, project_root
+    ):
+        try:
+            affected_presets = _installed_presets_affecting_agent(project_root, key)
+        except _PresetRegistryUnreadableError as exc:
+            console.print(
+                f"[red]Error:[/red] Cannot change '{key}' command layout: the "
+                f"preset registry could not be read to verify installed presets."
+            )
+            console.print(f"[dim]Details:[/dim] {_cli_error_detail(exc)}")
+            console.print(
+                "A layout change cannot reconcile preset artifacts, so the "
+                "migration is refused while the preset registry state is "
+                "unknown. Fix or restore "
+                "[cyan].specify/presets/.registry[/cyan] and retry."
+            )
+            raise typer.Exit(1)
+        if affected_presets:
+            preset_list = ", ".join(sorted(affected_presets))
+            console.print(
+                f"[red]Error:[/red] Cannot change '{key}' command layout while "
+                f"preset override(s) are installed: [bold]{preset_list}[/bold]."
+            )
+            console.print(
+                "Preset artifacts cannot yet be reconciled across a command↔skills "
+                "layout change, so the migration would orphan their files and leave "
+                "the preset registry inconsistent."
+            )
+            console.print(
+                "Remove the preset(s), run the upgrade, then reinstall them:\n"
+                f"  [cyan]specify preset remove <id>[/cyan]\n"
+                f"  [cyan]specify integration upgrade {key} "
+                f"--integration-options \"...\"[/cyan]\n"
+                f"  [cyan]specify preset add <id>[/cyan]"
+            )
+            raise typer.Exit(1)
+
     # Ensure shared infrastructure is up to date; --force overwrites existing files.
     infra_integration = integration
     infra_key = key
@@ -415,7 +554,8 @@ def integration_upgrade(
         selected_script,
         force=force,
         invoke_separator=_invoke_separator_for_integration(
-            infra_integration, current, infra_key, infra_parsed
+            infra_integration, current, infra_key, infra_parsed,
+            project_root=project_root,
         ),
     )
     if os.name != "nt":
@@ -441,6 +581,7 @@ def integration_upgrade(
             script_type=selected_script,
             raw_options=raw_options,
             parsed_options=parsed_options,
+            project_root=project_root,
         )
         if installed_key == key:
             try:
@@ -448,7 +589,8 @@ def integration_upgrade(
                     project_root,
                     selected_script,
                     invoke_separator=_invoke_separator_for_integration(
-                        integration, {"integration_settings": settings}, key, parsed_options
+                        integration, {"integration_settings": settings}, key, parsed_options,
+                        project_root=project_root,
                     ),
                     force=force,
                     refresh_managed=True,
@@ -463,7 +605,12 @@ def integration_upgrade(
         new_manifest.save()
         _write_integration_json(project_root, installed_key, installed_keys, settings)
         if installed_key == key:
-            _update_init_options_for_integration(project_root, integration, script_type=selected_script)
+            _update_init_options_for_integration(
+                project_root,
+                integration,
+                script_type=selected_script,
+                parsed_options=parsed_options,
+            )
         else:
             _refresh_init_options_speckit_version(project_root)
     except Exception as exc:
@@ -487,7 +634,15 @@ def integration_upgrade(
     if stale_keys:
         stale_manifest = IntegrationManifest(key, project_root, version="stale-cleanup")
         stale_manifest._files = {k: old_files[k] for k in stale_keys}
-        stale_removed, _ = stale_manifest.uninstall(project_root, force=True)
+        # remove_manifest=False: this throwaway manifest shares ``key`` with the
+        # real one just saved above (new_manifest.save()).  Letting uninstall()
+        # delete ``{key}.manifest.json`` would wipe the freshly-written manifest
+        # whenever an upgrade shrinks the tracked file set (e.g. Bob migrating
+        # from the legacy commands layout to skills), leaving the integration
+        # untracked and un-upgradeable.
+        stale_removed, _ = stale_manifest.uninstall(
+            project_root, force=True, remove_manifest=False
+        )
         if stale_removed:
             console.print(f"  Removed {len(stale_removed)} stale file(s) from previous install")
 
@@ -497,6 +652,55 @@ def integration_upgrade(
     # Done after the upgrade has fully settled (Phase 2 included) and outside
     # the try/except above so this best-effort step cannot affect upgrade
     # success.
+    #
+    # Layout-change reconciliation: a dual-mode agent (e.g. Bob) can flip
+    # between the legacy commands layout and the skills layout across an
+    # upgrade (``upgrade bob --integration-options "--skills"`` / reverse
+    # ``--legacy-commands``). Phase 2 above only removes stale files tracked by
+    # the *integration* manifest (core commands); extension artifacts are
+    # tracked separately in the extension registry, so the old layout's
+    # extension command/skill files would otherwise linger as orphans. When the
+    # layout actually changed, first unregister the agent's extension artifacts
+    # (removing old-layout files and clearing per-agent registry entries) so the
+    # re-registration below recreates them in the new layout. ``upgrade``s that
+    # don't change layout skip this to avoid needless remove/re-add churn.
+    #
+    # Only the *active* integration is reconciled this way (``installed_key ==
+    # key``).  ``ExtensionManager.unregister_agent_artifacts`` treats the
+    # per-extension ``registered_skills`` list as belonging to the passed agent
+    # and, when that agent's skills directory is absent, falls back to scanning
+    # every agent's skills directory — so running it for a *secondary*
+    # (non-active) agent could delete or untrack the *active* agent's extension
+    # skills.  The subsequent re-registration cannot repair that because
+    # extension skill rendering is intentionally scoped to the active agent
+    # (#2948).  Extension skills only ever exist for the active agent, so
+    # skipping the unregister for a secondary agent orphans nothing new: a
+    # secondary agent only has extension *command* files, which the
+    # re-registration below rewrites in place regardless of layout.
+    #
+    # Known limitation: preset command/skill artifacts are NOT reconciled on a
+    # layout change. There is no agent-scoped preset re-registration mechanism
+    # anywhere in the CLI — ``use`` / ``switch`` / ``upgrade`` never reconcile
+    # presets for any agent (presets are only (un)registered at preset
+    # install/remove time). Rather than silently orphan them, the guard near
+    # the top of this function rejects a layout-changing upgrade while preset
+    # overrides are installed, so control only reaches here (with a changed
+    # layout) when no preset artifacts are at stake. Full preset reconciliation
+    # would require a new cross-cutting PresetManager subsystem affecting every
+    # dual-layout agent, which is out of scope for this Bob migration.
+    if (
+        installed_key == key
+        and _manifest_tracks_skill_layout(old_manifest)
+        != _manifest_tracks_skill_layout(new_manifest)
+    ):
+        _unregister_extensions_for_agent(
+            project_root,
+            key,
+            continuing=(
+                "The integration layout changed, but old-layout extension "
+                "artifacts may need manual cleanup."
+            ),
+        )
     _register_extensions_for_agent(
         project_root,
         key,

src/specify_cli/integrations/base.py

@@ -14,6 +14,7 @@ Provides:
 from __future__ import annotations
 
 import os
+import platform
 import re
 import shlex
 import shutil
@@ -160,17 +161,66 @@ class IntegrationBase(ABC):
         return []
 
     def effective_invoke_separator(
-        self, parsed_options: dict[str, Any] | None = None
+        self,
+        parsed_options: dict[str, Any] | None = None,
+        project_root: Path | None = None,
     ) -> str:
         """Return the invoke separator for the given options.
 
         Subclasses whose separator depends on runtime options (e.g.
         Copilot in ``--skills`` mode) should override this method.
-        The default implementation ignores *parsed_options* and returns
-        the class-level ``invoke_separator``.
+        The default implementation ignores *parsed_options* and
+        *project_root* and returns the class-level ``invoke_separator``.
         """
         return self.invoke_separator
 
+    def invoke_separator_for_mode(self, skills_enabled: bool) -> str:
+        """Command-ref separator given the project's *resolved* skills state.
+
+        Registration paths (extension / preset command rendering) have no CLI
+        ``parsed_options`` — only the persisted ``ai_skills`` flag — so they
+        resolve the command-reference separator through this hook rather than
+        the static ``AGENT_CONFIGS[key]["invoke_separator"]`` value, which
+        cannot represent an agent whose separator differs between its skills
+        and command layouts.
+
+        The default is mode-independent and returns exactly what
+        ``_build_agent_configs`` would place in ``AGENT_CONFIGS`` (the
+        ``registrar_config`` override if present, else the class-level
+        ``invoke_separator``), so single-layout agents are unaffected.
+        Dual-mode agents whose separator depends on the layout (e.g. Bob:
+        ``-`` for skills, ``.`` for legacy commands) override this.
+        """
+        cfg = self.registrar_config or {}
+        return cfg.get("invoke_separator", self.invoke_separator)
+
+    def is_skills_mode(
+        self,
+        parsed_options: dict[str, Any] | None = None,
+        project_root: Path | None = None,
+    ) -> bool:
+        """Return whether this integration scaffolds skills for these options.
+
+        This is the single, well-defined hook the shared init/install/upgrade
+        machinery consults to decide whether to persist ``ai_skills=True`` and
+        render skill invocations.  It replaces ad-hoc ``isinstance`` /
+        ``getattr(self, "_skills_mode", ...)`` probing so an integration's
+        internal representation never has to leak into shared dispatch code.
+
+        *project_root* is optional context for the ``use`` / ``switch`` /
+        ``upgrade`` path, where no ``setup()`` runs and *parsed_options* may be
+        empty: dual-mode integrations can consult the already-installed
+        on-disk layout to avoid silently migrating an existing project to a
+        different mode.  The default ignores it.
+
+        The default (command-first integrations, e.g. Copilot's default
+        layout) is skills mode only when ``--skills`` was requested.
+        ``SkillsIntegration`` overrides this to return ``True`` by default;
+        skills-first integrations that expose a legacy opt-out (e.g. Bob)
+        override it to honor their own flag.
+        """
+        return bool((parsed_options or {}).get("skills"))
+
     def build_exec_args(
         self,
         prompt: str,
@@ -619,6 +669,46 @@ class IntegrationBase(ABC):
             return name
         return sys.executable or "python3"
 
+    @staticmethod
+    def build_python_invocation(
+        script_command: str, project_root: Path | None = None
+    ) -> str:
+        """Build a Python script command for the current platform shell."""
+        interpreter = IntegrationBase.resolve_python_interpreter(project_root)
+        if os.name == "nt" and not re.fullmatch(r"[A-Za-z0-9_./:\\-]+", interpreter):
+            quoted_interpreter = interpreter.replace("'", "''")
+            interpreter = f"& '{quoted_interpreter}'"
+        elif os.name != "nt":
+            interpreter = shlex.quote(interpreter)
+        return f"{interpreter} {script_command}"
+
+    @staticmethod
+    def select_script_variant(
+        requested: object, script_commands: dict[str, str]
+    ) -> str:
+        """Select the requested variant or a runnable platform fallback."""
+        if isinstance(requested, str) and requested in script_commands:
+            return requested
+
+        platform_variant = (
+            "ps" if platform.system().lower().startswith("win") else "sh"
+        )
+        secondary_variant = "sh" if platform_variant == "ps" else "ps"
+        fallbacks = (
+            (platform_variant, "py")
+            if requested == "py"
+            else (platform_variant, secondary_variant, "py")
+        )
+        for candidate in fallbacks:
+            if candidate in script_commands:
+                return candidate
+
+        available = ", ".join(sorted(script_commands)) or "none"
+        raise ValueError(
+            "No runnable script variant for this platform: "
+            f"requested {requested!r}; available: {available}"
+        )
+
     @staticmethod
     def _interpreter_runs(path: str) -> bool:
         """Return True when *path* executes as a Python interpreter.
@@ -653,7 +743,8 @@ class IntegrationBase(ABC):
         """Process a raw command template into agent-ready content.
 
         Performs the same transformations as the release script:
-        1. Extract ``scripts.<script_type>`` value from YAML frontmatter
+        1. Select ``scripts.<script_type>`` from YAML frontmatter, falling
+           back to a runnable platform shell or Python variant when unavailable
         2. Replace ``{SCRIPT}`` with the extracted script command
         3. Strip ``scripts:`` section from frontmatter
         4. Replace ``{ARGS}`` and ``$ARGUMENTS`` with *arg_placeholder*
@@ -662,37 +753,46 @@ class IntegrationBase(ABC):
         7. Replace ``__SPECKIT_COMMAND_<NAME>__`` with invocation strings
         """
         # 1. Extract script command from frontmatter
-        script_command = ""
-        script_pattern = re.compile(
-            rf"^\s*{re.escape(script_type)}:\s*(.+)$", re.MULTILINE
-        )
+        script_commands: dict[str, str] = {}
+        script_pattern = re.compile(r"^\s*([A-Za-z0-9_-]+):\s*(.+)$")
         # Find the scripts: block
+        in_frontmatter = False
         in_scripts = False
         for line in content.splitlines():
-            if line.strip() == "scripts:":
+            if line == "---":
+                if in_frontmatter:
+                    break
+                in_frontmatter = True
+                continue
+            if not in_frontmatter:
+                continue
+            if line == "scripts:":
                 in_scripts = True
                 continue
             if in_scripts and line and not line[0].isspace():
-                in_scripts = False
+                break
             if in_scripts:
                 m = script_pattern.match(line)
                 if m:
-                    script_command = m.group(1).strip()
-                    break
+                    script_commands[m.group(1)] = m.group(2).strip()
+
+        selected_script_type = (
+            IntegrationBase.select_script_variant(script_type, script_commands)
+            if script_commands
+            else ""
+        )
+
+        script_command = script_commands.get(selected_script_type, "")
 
         # 2. Replace {SCRIPT}
         if script_command:
             # For the Python script type, prefix the resolved interpreter so
             # the command is portable (``.py`` files are not directly
             # executable on Windows).
-            if script_type == "py":
-                interpreter = IntegrationBase.resolve_python_interpreter(project_root)
-                # Quote the interpreter if it contains whitespace (e.g. an
-                # absolute ``sys.executable`` path under Windows
-                # ``Program Files``) so it isn't split into multiple args.
-                if any(ch.isspace() for ch in interpreter):
-                    interpreter = f'"{interpreter}"'
-                script_command = f"{interpreter} {script_command}"
+            if selected_script_type == "py":
+                script_command = IntegrationBase.build_python_invocation(
+                    script_command, project_root
+                )
             content = content.replace("{SCRIPT}", script_command)
 
         # 3. Strip scripts: section from frontmatter
@@ -1376,6 +1476,14 @@ class SkillsIntegration(IntegrationBase):
 
     invoke_separator = "-"
 
+    def is_skills_mode(
+        self,
+        parsed_options: dict[str, Any] | None = None,
+        project_root: Path | None = None,
+    ) -> bool:
+        """Skills-native integrations scaffold skills unconditionally."""
+        return True
+
     def build_exec_args(
         self,
         prompt: str,

src/specify_cli/integrations/bob/__init__.py

@@ -1,10 +1,140 @@
-"""IBM Bob integration."""
+"""IBM Bob integration.
 
-from ..base import MarkdownIntegration
+Bob 2.0 uses the ``.bob/skills/speckit-<name>/SKILL.md`` layout by default.
+The legacy ``.bob/commands/*.md`` layout (Bob 1.x) remains available as an
+opt-in via ``--integration-options "--legacy-commands"``.
+
+Bob is a *dual-mode* integration: whether it scaffolds skills or commands is
+a per-project **configuration** decision (the ``--legacy-commands`` option,
+persisted as ``ai_skills`` in init-options), not a property of the class.
+It therefore extends :class:`IntegrationBase` (like Copilot, the other
+dual-mode agent) and resolves the mode through the ``is_skills_mode`` hook,
+delegating the actual scaffolding to a per-layout helper.
+
+Deprecation cycle:
+  This release:  Skills layout is the default; legacy ``.bob/commands/`` is
+                 opt-in via ``--legacy-commands``.
+  Next cycle:    ``--legacy-commands`` flag removed.
+"""
+
+from __future__ import annotations
+
+import warnings
+from pathlib import Path
+from typing import Any
+
+import typer
+
+from ..base import (
+    IntegrationBase,
+    IntegrationOption,
+    MarkdownIntegration,
+    SkillsIntegration,
+)
+from ..manifest import IntegrationManifest
 
 
-class BobIntegration(MarkdownIntegration):
+def _validate_mode_options(parsed_options: dict[str, Any] | None) -> None:
+    """Reject ``--skills`` and ``--legacy-commands`` used together.
+
+    The two flags select opposite layouts, so combining them is ambiguous.
+    Fail fast with the same clean exit-1 UX as other bad-option paths rather
+    than silently letting one win.
+    """
+    opts = parsed_options or {}
+    if opts.get("skills") and opts.get("legacy_commands"):
+        from ..._console import console
+
+        console.print(
+            "[red]Error:[/red] --skills and --legacy-commands are mutually "
+            "exclusive; pass only one."
+        )
+        raise typer.Exit(1)
+
+
+def _warn_legacy_commands_deprecated() -> None:
+    warnings.warn(
+        "Bob legacy commands mode (.bob/commands/) is deprecated and will be "
+        "removed in a future Spec Kit release. Omit --legacy-commands to use "
+        "the default skills layout (.bob/skills/).",
+        UserWarning,
+        stacklevel=3,
+    )
+
+
+class _BobSkillsHelper(SkillsIntegration):
+    """Default-mode helper: ``.bob/skills/speckit-<name>/SKILL.md``.
+
+    Not registered in the integration registry — used only as a delegate by
+    :class:`BobIntegration` for skills-mode ``setup()``.
+    """
+
     key = "bob"
+    config = {
+        "name": "IBM Bob",
+        "folder": ".bob/",
+        "commands_subdir": "skills",
+        "install_url": None,
+        "requires_cli": False,
+    }
+    registrar_config = {
+        "dir": ".bob/skills",
+        "format": "markdown",
+        "args": "$ARGUMENTS",
+        "extension": "/SKILL.md",
+    }
+
+    def post_process_skill_content(self, content: str) -> str:
+        """Bob skills are intent-activated; no slash-command note is needed."""
+        return content
+
+
+class _BobMarkdownHelper(MarkdownIntegration):
+    """Legacy-mode helper: ``.bob/commands/speckit.<name>.md`` (Bob 1.x).
+
+    Not registered in the integration registry — used only as a delegate by
+    :class:`BobIntegration` when ``--legacy-commands`` is passed.  Declares
+    ``invoke_separator="."`` so command-reference tokens render as Bob 1.x
+    ``/speckit.<name>`` invocations.
+    """
+
+    key = "bob"
+    invoke_separator = "."
+    config = {
+        "name": "IBM Bob",
+        "folder": ".bob/",
+        "commands_subdir": "commands",
+        "install_url": None,
+        "requires_cli": False,
+    }
+    registrar_config = {
+        "dir": ".bob/commands",
+        "format": "markdown",
+        "args": "$ARGUMENTS",
+        "extension": ".md",
+        "invoke_separator": ".",
+    }
+
+
+class BobIntegration(IntegrationBase):
+    """Integration for IBM Bob IDE (dual-mode; skills by default).
+
+    Whether a project uses the skills or the legacy commands layout is a
+    configuration choice resolved by :meth:`is_skills_mode`, not the class
+    hierarchy.  ``setup()`` delegates to the matching helper.
+
+    ``registrar_config`` mirrors the *commands* layout (``extension: ".md"``,
+    ``dir: ".bob/commands"``) — the same pattern Copilot uses — so that
+    ``CommandRegistrar.AGENT_CONFIGS["bob"]`` drives extension/preset
+    registration into ``.bob/commands/`` for legacy-mode projects, while
+    skills-mode projects have that command registration transparently skipped
+    (``skills_mode_active`` becomes ``True`` because ``ai_skills=True`` and
+    ``extension != "/SKILL.md"``) and receive extension skills instead.
+    ``invoke_separator = "-"`` matches the default (skills) layout.
+    """
+
+    key = "bob"
+    invoke_separator = "-"
     config = {
         "name": "IBM Bob",
         "folder": ".bob/",
@@ -18,3 +148,136 @@ class BobIntegration(MarkdownIntegration):
         "args": "$ARGUMENTS",
         "extension": ".md",
     }
+
+    @classmethod
+    def options(cls) -> list[IntegrationOption]:
+        return [
+            IntegrationOption(
+                "--skills",
+                is_flag=True,
+                default=False,
+                help=(
+                    "Force the default skills layout (.bob/skills/), overriding "
+                    "on-disk auto-detection. Use this to migrate a legacy "
+                    "commands install to skills, e.g. "
+                    "`integration upgrade bob --integration-options \"--skills\"`"
+                ),
+            ),
+            IntegrationOption(
+                "--legacy-commands",
+                is_flag=True,
+                default=False,
+                help=(
+                    "Scaffold commands as legacy .bob/commands/*.md files "
+                    "(Bob 1.x layout, deprecated) instead of the default "
+                    "skills layout"
+                ),
+            ),
+        ]
+
+    def is_skills_mode(
+        self,
+        parsed_options: dict[str, Any] | None = None,
+        project_root: Path | None = None,
+    ) -> bool:
+        """Bob is skills-first; ``--legacy-commands`` opts out.
+
+        Precedence:
+
+        1. Explicit ``--skills`` wins — it *forces* skills mode regardless of
+           what is already on disk.  This is the supported migration / opt-in
+           path: ``integration upgrade bob --integration-options "--skills"``
+           converts a legacy commands install to the skills layout (setup()
+           scaffolds ``.bob/skills`` and the upgrade's stale-file pass removes
+           the old ``.bob/commands`` files).
+        2. Explicit ``--legacy-commands`` opts out to the Bob 1.x layout.
+        3. Otherwise, when a *project_root* is supplied, the layout is inferred
+           from **managed Spec Kit artifacts** (see below).
+        4. A fresh project (no managed artifacts, no flags) defaults to skills.
+
+        The disk-detection fallback exists because on ``use`` / ``switch`` /
+        ``upgrade`` (without an explicit ``--skills`` / ``--legacy-commands``)
+        *parsed_options* is typically empty: no flag was passed, and existing
+        Bob 1.x installs never persisted a ``legacy_commands`` option to
+        recover.  This is independent of whether ``setup()`` runs — ``upgrade``
+        *does* call :meth:`setup` (see ``_migrate_commands.integration_upgrade``),
+        but it passes those same empty *parsed_options*, so without disk
+        detection the mode would resolve to the skills default.  Defaulting to
+        skills there would rewrite such a project's ``ai_skills`` flag to
+        ``True`` even though it still only contains a command layout, silently
+        switching its extension / command-reference handling.  So the layout is
+        inferred from managed Spec Kit artifacts, not the mere presence of a
+        ``.bob/skills/`` directory: a user may keep unrelated Bob 2 skills in
+        ``.bob/skills/`` while their Spec Kit commands still live in
+        ``.bob/commands/speckit.*.md``.  We therefore treat the project as
+        legacy (command) mode only when managed Spec Kit command files exist
+        and no managed Spec Kit skills (``speckit-*`` skill dirs) do.  Passing
+        ``--skills`` overrides this so users are never trapped in legacy mode.
+        """
+        opts = parsed_options or {}
+        _validate_mode_options(opts)
+        if opts.get("skills", False):
+            return True
+        if opts.get("legacy_commands", False):
+            return False
+        if project_root is not None:
+            bob_dir = Path(project_root) / ".bob"
+            has_managed_skills = any((bob_dir / "skills").glob("speckit-*"))
+            has_managed_commands = any((bob_dir / "commands").glob("speckit.*.md"))
+            if has_managed_commands and not has_managed_skills:
+                return False
+        return True
+
+    def effective_invoke_separator(
+        self,
+        parsed_options: dict[str, Any] | None = None,
+        project_root: Path | None = None,
+    ) -> str:
+        """``"."`` for the legacy commands layout, ``"-"`` for skills.
+
+        *project_root* lets the ``use`` / ``switch`` / ``upgrade`` path — which
+        refreshes shared infrastructure *before* persisting init-options —
+        detect an already-installed legacy layout, so core command references
+        are rendered with the correct separator instead of defaulting to the
+        skills ``-``.
+        """
+        return "-" if self.is_skills_mode(parsed_options, project_root) else "."
+
+    def invoke_separator_for_mode(self, skills_enabled: bool) -> str:
+        """Resolve the command-ref separator from a project's persisted mode.
+
+        Skills projects render ``/speckit-<cmd>``; legacy command projects
+        render Bob 1.x ``/speckit.<cmd>``.  Extension/preset registration
+        consults this (via the persisted ``ai_skills`` flag) so both layouts
+        get the correct separator despite sharing one static ``AGENT_CONFIGS``
+        entry.
+        """
+        return "-" if skills_enabled else "."
+
+    def post_process_skill_content(self, content: str) -> str:
+        """Bob skills are intent-activated; no slash-command note is injected.
+
+        Preset/extension skill generators call this on the *registered*
+        ``BobIntegration`` instance, not on :class:`_BobSkillsHelper`, so the
+        no-op must be repeated here (delegating to the helper) — otherwise
+        those paths would inherit ``IntegrationBase``'s default and inject
+        ``/speckit-*`` hook guidance that core Bob skills intentionally omit.
+        """
+        return _BobSkillsHelper().post_process_skill_content(content)
+
+    def setup(
+        self,
+        project_root: Path,
+        manifest: IntegrationManifest,
+        parsed_options: dict[str, Any] | None = None,
+        **opts: Any,
+    ) -> list[Path]:
+        parsed_options = parsed_options or {}
+        if self.is_skills_mode(parsed_options, project_root):
+            return _BobSkillsHelper().setup(
+                project_root, manifest, parsed_options, **opts
+            )
+        _warn_legacy_commands_deprecated()
+        return MarkdownIntegration.setup(
+            _BobMarkdownHelper(), project_root, manifest, parsed_options, **opts
+        )

src/specify_cli/integrations/catalog.py

@@ -40,6 +40,25 @@ class IntegrationDescriptorError(Exception):
     """Raised when an integration.yml descriptor is invalid."""
 
 
+def _catalog_shape_error(payload: Any) -> Optional[str]:
+    """Return a human-readable reason if *payload* is not a valid integration
+    catalog document, else ``None``.
+
+    Shared by the fresh-fetch and cache-read paths so both enforce the same
+    format contract: a JSON object carrying ``schema_version`` and a mapping
+    ``integrations``. Keeping a single validator prevents the two paths from
+    drifting (e.g. a cache that skips the ``schema_version`` check and lets an
+    older/poisoned payload bypass validation).
+    """
+    if not isinstance(payload, dict):
+        return "expected a JSON object"
+    if "schema_version" not in payload or "integrations" not in payload:
+        return "missing required 'schema_version' or 'integrations' key"
+    if not isinstance(payload.get("integrations"), dict):
+        return "'integrations' must be a JSON object"
+    return None
+
+
 # ---------------------------------------------------------------------------
 # IntegrationCatalogEntry
 # ---------------------------------------------------------------------------
@@ -153,7 +172,18 @@ class IntegrationCatalog(CatalogStackBase):
                     cached_at = cached_at.replace(tzinfo=timezone.utc)
                 age = (datetime.now(timezone.utc) - cached_at).total_seconds()
                 if age < self.CACHE_DURATION:
-                    return json.loads(cache_file.read_text(encoding="utf-8"))
+                    cached = json.loads(cache_file.read_text(encoding="utf-8"))
+                    # A poisoned/older-format cache must clear the SAME shape
+                    # contract as a fresh fetch (via the shared validator) —
+                    # otherwise a payload like [], {"integrations": []}, or one
+                    # missing "schema_version" is returned and later crashes on
+                    # .items()/.get() or silently bypasses the format contract.
+                    # The ValueError is caught just below, which drops the
+                    # corrupt cache and refetches from source.
+                    shape_error = _catalog_shape_error(cached)
+                    if shape_error is not None:
+                        raise ValueError(f"cached catalog has invalid shape: {shape_error}")
+                    return cached
             except (json.JSONDecodeError, ValueError, KeyError, TypeError, AttributeError, OSError, UnicodeError):
                 # Cache is invalid or stale metadata; delete and refetch from source.
                 try:
@@ -172,20 +202,10 @@ class IntegrationCatalog(CatalogStackBase):
                     self._validate_catalog_url(final_url)
                 catalog_data = json.loads(resp.read())
 
-            if not isinstance(catalog_data, dict):
+            shape_error = _catalog_shape_error(catalog_data)
+            if shape_error is not None:
                 raise IntegrationCatalogError(
-                    f"Invalid catalog format from {entry.url}: expected a JSON object"
-                )
-            if (
-                "schema_version" not in catalog_data
-                or "integrations" not in catalog_data
-            ):
-                raise IntegrationCatalogError(
-                    f"Invalid catalog format from {entry.url}"
-                )
-            if not isinstance(catalog_data.get("integrations"), dict):
-                raise IntegrationCatalogError(
-                    f"Invalid catalog format from {entry.url}: 'integrations' must be a JSON object"
+                    f"Invalid catalog format from {entry.url}: {shape_error}"
                 )
 
             try:
@@ -429,7 +449,8 @@ class IntegrationCatalog(CatalogStackBase):
                     )
                 try:
                     normalized_priority = int(raw_priority)
-                except (TypeError, ValueError):
+                except (TypeError, ValueError, OverflowError):
+                    # OverflowError: int(float("inf")) — a ``priority: .inf``.
                     raise IntegrationValidationError(
                         f"Invalid catalog entry at index {idx} in {config_path}: "
                         f"'priority' must be an integer, got "
@@ -537,7 +558,8 @@ class IntegrationCatalog(CatalogStackBase):
             else:
                 try:
                     priority = int(raw_priority)
-                except (TypeError, ValueError):
+                except (TypeError, ValueError, OverflowError):
+                    # OverflowError: int(float("inf")) — a ``priority: .inf``.
                     priority = yaml_idx + 1
             priority_pairs.append((priority, yaml_idx))
         if not priority_pairs:

src/specify_cli/integrations/cline/__init__.py

@@ -77,6 +77,19 @@ class ClineIntegration(MarkdownIntegration):
         """Cline uses hyphenated filenames (e.g. speckit-git-commit.md)."""
         return format_cline_command_name(template_name) + ".md"
 
+    def build_command_invocation(self, command_name: str, args: str = "") -> str:
+        """Cline installs hyphenated slash-commands (``/speckit-<name>``), so the
+        dispatch invocation must match. The inherited MarkdownIntegration default
+        builds the dotted ``/speckit.<name>``, which references a command Cline
+        never registered. Reuse the same hyphenation as command_filename /
+        the injected frontmatter name (see ``format_cline_command_name``),
+        mirroring the forge integration.
+        """
+        invocation = "/" + format_cline_command_name(command_name)
+        if args:
+            invocation = f"{invocation} {args}"
+        return invocation
+
     def process_template(self, *args, **kwargs):
         """Ensure shared templates render Cline command references with hyphens."""
         kwargs.setdefault("invoke_separator", self.invoke_separator)

src/specify_cli/integrations/copilot/__init__.py

@@ -122,7 +122,9 @@ class CopilotIntegration(IntegrationBase):
     _skills_mode: bool = False
 
     def effective_invoke_separator(
-        self, parsed_options: dict[str, Any] | None = None
+        self,
+        parsed_options: dict[str, Any] | None = None,
+        project_root: Path | None = None,
     ) -> str:
         """Return ``"-"`` when skills mode is requested, ``"."`` otherwise."""
         if parsed_options and parsed_options.get("skills"):
@@ -131,6 +133,33 @@ class CopilotIntegration(IntegrationBase):
             return "-"
         return self.invoke_separator
 
+    def is_skills_mode(
+        self,
+        parsed_options: dict[str, Any] | None = None,
+        project_root: Path | None = None,
+    ) -> bool:
+        """Copilot is skills mode when ``--skills`` was requested.
+
+        On the init path ``setup()`` has already recorded the choice in
+        ``self._skills_mode``; on the ``use``/``install`` path (where no
+        ``setup()`` runs) the signal comes from *parsed_options* (#3550), which
+        round-trips because ``--skills`` is persisted in the stored options.
+        """
+        if parsed_options and parsed_options.get("skills"):
+            return True
+        return self._skills_mode
+
+    def invoke_separator_for_mode(self, skills_enabled: bool) -> str:
+        """Skills projects render ``/speckit-<cmd>``; default markdown ``.``.
+
+        Copilot is dual-layout, so — like Bob — the command-reference
+        separator depends on the persisted ``ai_skills`` state rather than a
+        single static value.  This keeps preset/extension command refs in a
+        Copilot skills project consistent with ``build_command_invocation``
+        (which emits ``/speckit-<stem>``).
+        """
+        return "-" if skills_enabled else self.invoke_separator
+
     @classmethod
     def options(cls) -> list[IntegrationOption]:
         return [

src/specify_cli/integrations/droid/__init__.py

@@ -0,0 +1,135 @@
+"""Factory Droid CLI integration — skills-based agent.
+
+Droid discovers project skills from
+``.factory/skills/speckit-<name>/SKILL.md``. Spec Kit installs into that
+native tree so the generated skills are visible to Droid without extra
+configuration.
+
+See: https://docs.factory.ai/cli/configuration/skills
+"""
+
+from __future__ import annotations
+
+from ..base import SkillsIntegration
+
+
+class DroidIntegration(SkillsIntegration):
+    """Integration for Factory Droid CLI."""
+
+    key = "droid"
+    config = {
+        "name": "Factory Droid",
+        "folder": ".factory/",
+        "commands_subdir": "skills",
+        "install_url": "https://docs.factory.ai/cli/getting-started/overview",
+        "requires_cli": True,
+    }
+    registrar_config = {
+        "dir": ".factory/skills",
+        "format": "markdown",
+        "args": "$ARGUMENTS",
+        "extension": "/SKILL.md",
+    }
+    multi_install_safe = True
+
+    @staticmethod
+    def _inject_frontmatter_flag(content: str, key: str, value: str = "true") -> str:
+        """Insert ``key: value`` before the closing ``---`` if not already present.
+
+        Mirrors the helper used by ``ClaudeIntegration`` / ``VibeIntegration``
+        so per-agent frontmatter injection stays consistent across skills-based
+        integrations. Pre-scans for the key to keep injection idempotent.
+        """
+        lines = content.splitlines(keepends=True)
+
+        # Pre-scan: bail out if already present in frontmatter
+        dash_count = 0
+        for line in lines:
+            stripped = line.rstrip("\n\r")
+            if stripped == "---":
+                dash_count += 1
+                if dash_count == 2:
+                    break
+                continue
+            if dash_count == 1 and stripped.startswith(f"{key}:"):
+                return content
+
+        # Inject before the closing --- of frontmatter. Always emit a
+        # newline after the injected key so the key and the closing ---
+        # stay on separate lines even when the closing delimiter is the
+        # last line of the file with no trailing newline.
+        out: list[str] = []
+        dash_count = 0
+        injected = False
+        for line in lines:
+            stripped = line.rstrip("\n\r")
+            if stripped == "---":
+                dash_count += 1
+                if dash_count == 2 and not injected:
+                    out.append(f"{key}: {value}\n")
+                    injected = True
+            out.append(line)
+        return "".join(out)
+
+    def post_process_skill_content(self, content: str) -> str:
+        """Inject Droid-specific skill frontmatter flags.
+
+        Applies the shared hook-command normalization note (skills agents use
+        hyphenated ``/speckit-<name>`` invocations, not dotted ``/speckit.<name>``)
+        and the Droid-specific ``user-invocable`` / ``disable-model-invocation``
+        frontmatter flags so skills are both user- and Droid-invocable.
+        """
+        updated = super().post_process_skill_content(content)
+        updated = self._inject_frontmatter_flag(updated, "user-invocable")
+        updated = self._inject_frontmatter_flag(updated, "disable-model-invocation", "false")
+        return updated
+
+    def build_exec_args(
+        self,
+        prompt: str,
+        *,
+        model: str | None = None,
+        output_json: bool = True,
+    ) -> list[str] | None:
+        """Build CLI arguments for non-interactive ``droid`` execution.
+
+        Uses ``droid exec "<prompt>"`` for headless dispatch. Spec Kit does
+        not auto-apply any permission-bypass flag: operators who want to
+        skip interactive confirmation can pass it through
+        ``SPECKIT_INTEGRATION_DROID_EXTRA_ARGS`` (e.g.
+        ``SPECKIT_INTEGRATION_DROID_EXTRA_ARGS="--skip-permissions-unsafe"``).
+
+        Output format and model selection mirror the documented CLI flags:
+        ``--output-format json`` (when ``output_json`` is set) and
+        ``--model <id>``. Operator-supplied extra args via
+        ``SPECKIT_INTEGRATION_DROID_EXTRA_ARGS`` are appended after the
+        canonical Spec Kit flags so the canonical flags are guaranteed to
+        be present in argv. Note that with duplicate-flag CLI parsing the
+        later (operator-supplied) value may take precedence over the
+        canonical one, so operators can still override ``--model`` or
+        ``--output-format``.
+        """
+        if not self.config or not self.config.get("requires_cli"):
+            return None
+        args = [
+            self._resolve_executable(),
+            "exec",
+            prompt,
+        ]
+        # Operator-injected extra args are appended after Spec Kit's
+        # canonical --model / --output-format flags so the canonical
+        # flags are guaranteed to be present in argv regardless of
+        # whatever the operator passes via SPECKIT_INTEGRATION_DROID_EXTRA_ARGS.
+        # This is a deliberate inversion of the cursor-agent / opencode /
+        # codex ordering (which all apply extra args first, then append
+        # canonical flags so the canonical values win under duplicate-flag
+        # parsing). For Droid the canonical flag values are written into
+        # argv first, then the operator-supplied values follow; with
+        # duplicate-flag parsing the later (operator) value may therefore
+        # take precedence.
+        if model:
+            args.extend(["--model", model])
+        if output_json:
+            args.extend(["--output-format", "json"])
+        self._apply_extra_args_env_var(args)
+        return args

src/specify_cli/integrations/forge/__init__.py

@@ -91,6 +91,18 @@ class ForgeIntegration(MarkdownIntegration):
     }
     invoke_separator = "-"
 
+    def build_command_invocation(self, command_name: str, args: str = "") -> str:
+        """Forge installs hyphenated slash-commands (``/speckit-<name>``), so the
+        dispatch invocation must match. The inherited MarkdownIntegration default
+        builds the dotted ``/speckit.<name>``, which references a command Forge
+        never registered. Reuse the same hyphenation as the installed frontmatter
+        ``name`` (see ``format_forge_command_name``), mirroring the skills agents.
+        """
+        invocation = "/" + format_forge_command_name(command_name)
+        if args:
+            invocation = f"{invocation} {args}"
+        return invocation
+
     def setup(
         self,
         project_root: Path,

src/specify_cli/integrations/manifest.py

@@ -327,12 +327,18 @@ class IntegrationManifest:
         project_root: Path | None = None,
         *,
         force: bool = False,
+        remove_manifest: bool = True,
     ) -> tuple[list[Path], list[Path]]:
         """Remove tracked files whose hash still matches.
 
         Parameters:
-            project_root: Override for the project root.
-            force:        If ``True``, remove files even if modified.
+            project_root:    Override for the project root.
+            force:           If ``True``, remove files even if modified.
+            remove_manifest: If ``True`` (default), also delete this
+                integration's ``{key}.manifest.json``. Set ``False`` for
+                *partial* cleanups (e.g. the upgrade stale-file pass, which
+                builds a throwaway manifest over a subset of files) so the
+                real, freshly-saved manifest for the same key is not destroyed.
 
         Returns:
             ``(removed, skipped)`` — absolute paths.
@@ -393,7 +399,7 @@ class IntegrationManifest:
 
         # Remove the manifest file itself
         manifest = root / ".specify" / "integrations" / f"{self.key}.manifest.json"
-        if manifest.exists():
+        if remove_manifest and manifest.exists():
             manifest.unlink()
             parent = manifest.parent
             while parent != root:

src/specify_cli/presets/__init__.py

@@ -1171,7 +1171,7 @@ class PresetManager:
                             selected_ai, fm, body, self.project_root
                         )
                         body = self._resolve_skill_command_refs(
-                            body, registrar, selected_ai
+                            body, registrar, selected_ai, self.project_root
                         )
                     from ..integrations import get_integration
                     integration = get_integration(selected_ai) if isinstance(selected_ai, str) else None
@@ -1252,7 +1252,10 @@ class PresetManager:
 
     @staticmethod
     def _resolve_skill_command_refs(
-        body: str, registrar: "CommandRegistrar", selected_ai: str
+        body: str,
+        registrar: "CommandRegistrar",
+        selected_ai: str,
+        project_root: "Path | None" = None,
     ) -> str:
         """Render ``__SPECKIT_COMMAND_*__`` tokens in a skill body as invocations.
 
@@ -1261,10 +1264,30 @@ class PresetManager:
         slash-command invocation — ``/speckit-<cmd>`` for a ``-`` separator,
         ``/speckit.<cmd>`` for ``.`` — the same rendering the command layer
         applies via ``CommandRegistrar.register_commands()``.
+
+        For dual-layout agents (e.g. Bob) the separator depends on the
+        project's persisted skills state, so — when *project_root* is provided
+        — the separator is resolved from the integration via
+        ``invoke_separator_for_mode`` rather than the single static
+        ``AGENT_CONFIGS`` value.
         """
-        separator = registrar.AGENT_CONFIGS.get(selected_ai, {}).get(
-            "invoke_separator", "."
-        )
+        separator = None
+        if project_root is not None and isinstance(selected_ai, str):
+            try:
+                from .. import load_init_options
+                from ..integrations import get_integration
+
+                integration = get_integration(selected_ai)
+                if integration is not None:
+                    separator = integration.invoke_separator_for_mode(
+                        is_ai_skills_enabled(load_init_options(project_root))
+                    )
+            except Exception:
+                separator = None
+        if separator is None:
+            separator = registrar.AGENT_CONFIGS.get(selected_ai, {}).get(
+                "invoke_separator", "."
+            )
         return IntegrationBase.resolve_command_refs(body, separator)
 
     def _build_extension_skill_restore_index(self) -> Dict[str, Dict[str, Any]]:
@@ -1445,7 +1468,7 @@ class PresetManager:
             body = registrar.resolve_skill_placeholders(
                 selected_ai, frontmatter, body, self.project_root
             )
-            body = self._resolve_skill_command_refs(body, registrar, selected_ai)
+            body = self._resolve_skill_command_refs(body, registrar, selected_ai, self.project_root)
 
             for target_skill_name in target_skill_names:
                 skill_subdir = skills_dir / target_skill_name
@@ -1540,7 +1563,7 @@ class PresetManager:
                         selected_ai, frontmatter, body, self.project_root
                     )
                     body = self._resolve_skill_command_refs(
-                        body, registrar, selected_ai
+                        body, registrar, selected_ai, self.project_root
                     )
 
                 original_desc = frontmatter.get("description", "")
@@ -1592,7 +1615,7 @@ class PresetManager:
                         selected_ai, frontmatter, body, self.project_root
                     )
                     body = self._resolve_skill_command_refs(
-                        body, registrar, selected_ai
+                        body, registrar, selected_ai, self.project_root
                     )
 
                 command_name = extension_restore["command_name"]
@@ -2108,13 +2131,22 @@ class PresetCatalog:
         url: str,
         timeout: int = 10,
         extra_headers: Optional[Dict[str, str]] = None,
+        redirect_validator=None,
     ):
         """Open a URL with provider-based auth, trying each configured provider.
 
         Delegates to :func:`specify_cli.authentication.http.open_url`.
+        *redirect_validator*, when provided, is invoked as ``(old_url, new_url)``
+        before EACH redirect hop, so an HTTPS host guarantee can be enforced on
+        every intermediate URL, not just the terminal one.
         """
         from specify_cli.authentication.http import open_url
-        return open_url(url, timeout, extra_headers=extra_headers)
+        return open_url(
+            url,
+            timeout,
+            extra_headers=extra_headers,
+            redirect_validator=redirect_validator,
+        )
 
     def _resolve_github_release_asset_api_url(
         self,
@@ -2235,7 +2267,10 @@ class PresetCatalog:
                 )
             try:
                 priority = int(raw_priority)
-            except (TypeError, ValueError):
+            except (TypeError, ValueError, OverflowError):
+                # OverflowError: int(float("inf")) — a YAML ``priority: .inf``
+                # would otherwise escape as an uncaught traceback instead of the
+                # clean validation error (mirrors catalogs.py).
                 raise PresetValidationError(
                     f"Invalid priority for catalog '{item.get('name', idx + 1)}': "
                     f"expected integer, got {raw_priority!r}"
@@ -2401,7 +2436,21 @@ class PresetCatalog:
                 pass
 
         try:
-            with self._open_url(entry.url, timeout=10) as response:
+            # Validate EVERY redirect hop (not just the terminal URL): an
+            # https -> http -> attacker-controlled-https chain would pass a
+            # final-URL-only check while the insecure intermediate hop lets a
+            # network attacker rewrite the next redirect. redirect_validator runs
+            # before each hop; the final geturl() check is retained as a
+            # belt-and-braces guard. Mirrors bundler/services/adapters.py.
+            def _validate_redirect(_old_url: str, new_url: str) -> None:
+                self._validate_catalog_url(new_url)
+
+            with self._open_url(
+                entry.url, timeout=10, redirect_validator=_validate_redirect
+            ) as response:
+                final_url = response.geturl()
+                if final_url != entry.url:
+                    self._validate_catalog_url(final_url)
                 catalog_data = json.loads(response.read())
 
             self._validate_catalog_payload(catalog_data, entry.url)
@@ -2552,7 +2601,18 @@ class PresetCatalog:
                 pass
 
         try:
-            with self._open_url(catalog_url, timeout=10) as response:
+            # Same redirect hardening as _fetch_single_catalog: validate every
+            # redirect hop AND the final URL so this legacy single-catalog path
+            # is not vulnerable to an HTTPS->HTTP redirected payload either.
+            def _validate_redirect(_old_url: str, new_url: str) -> None:
+                self._validate_catalog_url(new_url)
+
+            with self._open_url(
+                catalog_url, timeout=10, redirect_validator=_validate_redirect
+            ) as response:
+                final_url = response.geturl()
+                if final_url != catalog_url:
+                    self._validate_catalog_url(final_url)
                 catalog_data = json.loads(response.read())
 
             # Validate catalog structure. Reuses the same helper as
@@ -2717,8 +2777,20 @@ class PresetCatalog:
 
         from urllib.parse import urlparse
 
-        parsed = urlparse(download_url)
-        is_localhost = parsed.hostname in ("localhost", "127.0.0.1", "::1")
+        # A malformed authority (e.g. an unterminated IPv6 bracket
+        # "https://[::1") makes urlparse / hostname access raise ValueError.
+        # The download_url comes from catalog payload data, so surface a clean
+        # PresetError rather than leaking a raw ValueError past the command
+        # handler (which only catches PresetError). Mirrors catalogs (#3435)
+        # and workflows/catalog.py (#3484).
+        try:
+            parsed = urlparse(download_url)
+            hostname = parsed.hostname
+        except ValueError:
+            raise PresetError(
+                f"Preset download URL is malformed: {download_url}"
+            ) from None
+        is_localhost = hostname in ("localhost", "127.0.0.1", "::1")
         if parsed.scheme != "https" and not (
             parsed.scheme == "http" and is_localhost
         ):

src/specify_cli/presets/_commands.py

@@ -13,6 +13,7 @@ from pathlib import Path
 
 import typer
 import yaml
+from rich.markup import escape as _escape_markup
 
 from .._console import console
 
@@ -107,8 +108,6 @@ def preset_add(
             try:
                 _parsed = _urlparse(from_url)
             except ValueError:
-                from rich.markup import escape as _escape_markup
-
                 console.print(f"[red]Error:[/red] Invalid URL: {_escape_markup(from_url)}")
                 raise typer.Exit(1)
 
@@ -141,9 +140,7 @@ def preset_add(
                 )
                 raise typer.Exit(1)
 
-            from rich.markup import escape as _esc
-
-            console.print(f"Installing preset from [cyan]{_esc(from_url)}[/cyan]...")
+            console.print(f"Installing preset from [cyan]{_escape_markup(from_url)}[/cyan]...")
             import urllib.error
             import tempfile
             import shutil
@@ -183,7 +180,7 @@ def preset_add(
                             except TypeError:
                                 output.write(response.read())
                 except urllib.error.URLError as e:
-                    console.print(f"[red]Error:[/red] Failed to download: {e}")
+                    console.print(f"[red]Error:[/red] Failed to download: {_escape_markup(str(e))}")
                     raise typer.Exit(1)
 
                 manifest = manager.install_from_zip(zip_path, speckit_version, priority)
@@ -240,13 +237,13 @@ def preset_add(
             raise typer.Exit(1)
 
     except PresetCompatibilityError as e:
-        console.print(f"[red]Compatibility Error:[/red] {e}")
+        console.print(f"[red]Compatibility Error:[/red] {_escape_markup(str(e))}")
         raise typer.Exit(1)
     except PresetValidationError as e:
-        console.print(f"[red]Validation Error:[/red] {e}")
+        console.print(f"[red]Validation Error:[/red] {_escape_markup(str(e))}")
         raise typer.Exit(1)
     except PresetError as e:
-        console.print(f"[red]Error:[/red] {e}")
+        console.print(f"[red]Error:[/red] {_escape_markup(str(e))}")
         raise typer.Exit(1)
 
 
@@ -288,7 +285,7 @@ def preset_search(
     try:
         results = catalog.search(query=query, tag=tag, author=author)
     except PresetError as e:
-        console.print(f"[red]Error:[/red] {e}")
+        console.print(f"[red]Error:[/red] {_escape_markup(str(e))}")
         raise typer.Exit(1)
 
     if not results:
@@ -582,7 +579,7 @@ def preset_catalog_list():
     try:
         active_catalogs = catalog.get_active_catalogs()
     except PresetValidationError as e:
-        console.print(f"[red]Error:[/red] {e}")
+        console.print(f"[red]Error:[/red] {_escape_markup(str(e))}")
         raise typer.Exit(1)
 
     console.print("\n[bold cyan]Active Preset Catalogs:[/bold cyan]\n")
@@ -647,7 +644,7 @@ def preset_catalog_add(
     try:
         tmp_catalog._validate_catalog_url(url)
     except PresetValidationError as e:
-        console.print(f"[red]Error:[/red] {e}")
+        console.print(f"[red]Error:[/red] {_escape_markup(str(e))}")
         raise typer.Exit(1)
 
     config_path = specify_dir / "preset-catalogs.yml"
@@ -658,7 +655,7 @@ def preset_catalog_add(
             config = yaml.safe_load(config_path.read_text(encoding="utf-8")) or {}
         except Exception as e:
             config_label = _display_project_path(project_root, config_path)
-            console.print(f"[red]Error:[/red] Failed to read {config_label}: {e}")
+            console.print(f"[red]Error:[/red] Failed to read {_escape_markup(str(config_label))}: {_escape_markup(str(e))}")
             raise typer.Exit(1)
     else:
         config = {}

src/specify_cli/shared_infra.py

@@ -402,8 +402,13 @@ def install_shared_infra(
     # Track every shared path the current bundle produces so we can detect
     # manifest entries the core no longer ships (stale-script cleanup, #3076).
     seen_rels: set[str] = set()
-    scripts_scanned = False
-    variant_dir = {"sh": "bash", "py": "python"}.get(script_type, "powershell")
+    scanned_variant_dirs: set[str] = set()
+    shell_variant = "powershell" if os.name == "nt" else "bash"
+    variant_dirs = (
+        ("python", shell_variant)
+        if script_type == "py"
+        else ("bash" if script_type == "sh" else "powershell",)
+    )
 
     def _decide_overwrite(rel: str, dst: Path) -> tuple[bool, str | None]:
         """Return (write, bucket) where bucket is 'skip', 'preserved', or None."""
@@ -458,69 +463,69 @@ def install_shared_infra(
     if scripts_src.is_dir():
         dest_scripts = project_path / ".specify" / "scripts"
         if _ensure_or_bucket_dir(dest_scripts):
-            variant_src = scripts_src / variant_dir
-            if variant_src.is_dir():
+            for variant_dir in variant_dirs:
+                variant_src = scripts_src / variant_dir
+                if not variant_src.is_dir():
+                    continue
                 dest_variant = dest_scripts / variant_dir
-                if _ensure_or_bucket_dir(dest_variant):
-                    for src_path in variant_src.rglob("*"):
-                        if not src_path.is_file():
-                            continue
-                        # Python bytecode caches are local artifacts, not
-                        # workflow scripts — never install them.
-                        if "__pycache__" in src_path.parts:
-                            continue
-                        # Mark scanned only once a real source file is seen. An
-                        # empty (or symlink-skipped) variant keeps this False, so
-                        # stale-cleanup is skipped — otherwise it would treat every
-                        # tracked script as obsolete and delete it. (The safety
-                        # hinge is this flag, not ``seen_rels``, which also holds
-                        # template paths populated later.)
-                        scripts_scanned = True
+                if not _ensure_or_bucket_dir(dest_variant):
+                    continue
+                for src_path in variant_src.rglob("*"):
+                    if not src_path.is_file():
+                        continue
+                    # Python bytecode caches are local artifacts, not
+                    # workflow scripts — never install them.
+                    if "__pycache__" in src_path.parts:
+                        continue
+                    # Mark scanned only once a real source file is seen. An
+                    # empty (or symlink-skipped) variant stays untracked, so
+                    # stale-cleanup cannot treat its managed scripts as obsolete.
+                    scanned_variant_dirs.add(variant_dir)
 
-                        rel_path = src_path.relative_to(variant_src)
-                        dst_path = dest_variant / rel_path
-                        rel = dst_path.relative_to(project_path).as_posix()
-                        seen_rels.add(rel)
-                        if not _safe_dest_or_bucket(dst_path, rel, parent_must_exist=False):
-                            continue
-                        write, bucket = _decide_overwrite(rel, dst_path)
-                        if not write:
-                            if bucket == "preserved":
-                                preserved_user_files.append(rel)
-                            else:
-                                skipped_files.append(rel)
-                                # Record the existing-on-disk file in the manifest so a
-                                # fresh manifest run against an already-populated
-                                # ``.specify/`` tree does not silently drop it (#2107).
-                                # ``prior_hashes`` is the function-scope snapshot taken
-                                # at entry, so this membership check is O(1) and avoids
-                                # the repeated ``dict(self._files)`` copy that
-                                # ``manifest.files`` performs on every access.
-                                if dst_path.is_file() and rel not in prior_hashes:
-                                    try:
-                                        manifest.record_existing(rel, recovered=True)
-                                    except (OSError, ValueError) as exc:
-                                        # Tolerate races / permission issues / non-file
-                                        # collisions so one weird path does not abort
-                                        # the whole install.
-                                        console.print(
-                                            f"[yellow]⚠[/yellow]  could not record {rel} in manifest: {exc}"
-                                        )
-                            continue
+                    rel_path = src_path.relative_to(variant_src)
+                    dst_path = dest_variant / rel_path
+                    rel = dst_path.relative_to(project_path).as_posix()
+                    seen_rels.add(rel)
+                    if not _safe_dest_or_bucket(dst_path, rel, parent_must_exist=False):
+                        continue
+                    write, bucket = _decide_overwrite(rel, dst_path)
+                    if not write:
+                        if bucket == "preserved":
+                            preserved_user_files.append(rel)
+                        else:
+                            skipped_files.append(rel)
+                            # Record the existing-on-disk file in the manifest so a
+                            # fresh manifest run against an already-populated
+                            # ``.specify/`` tree does not silently drop it (#2107).
+                            # ``prior_hashes`` is the function-scope snapshot taken
+                            # at entry, so this membership check is O(1) and avoids
+                            # the repeated ``dict(self._files)`` copy that
+                            # ``manifest.files`` performs on every access.
+                            if dst_path.is_file() and rel not in prior_hashes:
+                                try:
+                                    manifest.record_existing(rel, recovered=True)
+                                except (OSError, ValueError) as exc:
+                                    # Tolerate races / permission issues / non-file
+                                    # collisions so one weird path does not abort
+                                    # the whole install.
+                                    console.print(
+                                        f"[yellow]⚠[/yellow]  could not record {rel} in manifest: {exc}"
+                                    )
+                        continue
 
-                        if not _ensure_or_bucket_dir(dst_path.parent):
-                            continue
-                        content = src_path.read_text(encoding="utf-8")
-                        content = IntegrationBase.resolve_command_refs(content, invoke_separator)
-                        content = _resolve_dynamic_command_refs(content, invoke_separator)
-                        planned_copies.append(
-                            (
-                                dst_path,
-                                rel,
-                                content.encode("utf-8"),
-                                src_path.stat().st_mode & 0o777,
-                            )
+                    if not _ensure_or_bucket_dir(dst_path.parent):
+                        continue
+                    content = src_path.read_text(encoding="utf-8")
+                    content = IntegrationBase.resolve_command_refs(content, invoke_separator)
+                    content = _resolve_dynamic_command_refs(content, invoke_separator)
+                    planned_copies.append(
+                        (
+                            dst_path,
+                            rel,
+                            content.encode("utf-8"),
+                            src_path.stat().st_mode & 0o777,
                         )
+                    )
 
     templates_src = shared_templates_source(core_pack=core_pack, repo_root=repo_root)
     if templates_src.is_dir():
@@ -618,14 +623,16 @@ def install_shared_infra(
     # agent-context extension. Left behind, such an orphan can crash when it
     # sources a refreshed ``common.sh`` (#3076). Only run when the script source
     # was actually scanned (so a missing/empty source never triggers mass
-    # deletion), scoped to the active variant, and only for *managed* copies —
+    # deletion), scoped to the selected variants, and only for *managed* copies —
     # a user-customized file (hash diverges), a symlink, or a recovered entry is
     # preserved by ``_is_managed``.
-    if scripts_scanned:
+    if scanned_variant_dirs:
         stale_removed: list[str] = []
-        script_prefix = f".specify/scripts/{variant_dir}/"
+        script_prefixes = tuple(
+            f".specify/scripts/{variant_dir}/" for variant_dir in scanned_variant_dirs
+        )
         for rel in list(prior_hashes):
-            if rel in seen_rels or not rel.startswith(script_prefix):
+            if rel in seen_rels or not rel.startswith(script_prefixes):
                 continue
             # Guard corrupted/hand-edited manifest keys BEFORE any filesystem
             # access: absolute, ``..``, or (on Windows) drive-relative keys such

src/specify_cli/workflows/_commands.py

@@ -49,6 +49,13 @@ workflow_step_catalog_app = typer.Typer(
 )
 workflow_step_app.add_typer(workflow_step_catalog_app, name="catalog")
 
+workflow_overlay_app = typer.Typer(
+    name="overlay",
+    help="Manage workflow overlays",
+    add_completion=False,
+)
+workflow_app.add_typer(workflow_overlay_app, name="overlay")
+
 
 def _error_console(json_output: bool):
     """Console for error text: stderr under ``--json`` so the JSON stdout
@@ -192,6 +199,10 @@ def _reject_unsafe_workflow_storage(project_root: Path) -> None:
         project_root / ".specify" / "workflows" / "runs",
         ".specify/workflows/runs",
     )
+    _reject_unsafe_dir(
+        project_root / ".specify" / "workflows" / "overlays",
+        ".specify/workflows/overlays",
+    )
 
 
 def _scan_for_workflow_owner(parts: tuple[str, ...]) -> int | None:
@@ -366,7 +377,7 @@ def _resolve_installed_workflow_ownership(
 
 
 _WORKFLOW_ID_PATTERN = re.compile(r"^[a-z0-9](?:[a-z0-9-]*[a-z0-9])?$")
-_RESERVED_WORKFLOW_IDS: frozenset[str] = frozenset({"runs", "steps"})
+_RESERVED_WORKFLOW_IDS: frozenset[str] = frozenset({"overlays", "runs", "steps"})
 
 
 def _reject_insecure_download_redirect(old_url: str, new_url: str) -> None:
@@ -1262,14 +1273,18 @@ def workflow_status(
     engine = WorkflowEngine(project_root)
 
     if run_id:
+        # Route errors to stderr under --json so the stdout JSON stream stays
+        # parseable (mirrors `workflow run`/`workflow resume`); both handlers
+        # fire before the json_output branch below.
+        err = _error_console(json_output)
         try:
             from .engine import RunState
             state = RunState.load(run_id, project_root)
         except FileNotFoundError:
-            console.print(f"[red]Error:[/red] Run not found: {run_id}")
+            err.print(f"[red]Error:[/red] Run not found: {run_id}")
             raise typer.Exit(1)
         except ValueError as exc:
-            console.print(f"[red]Error:[/red] {_escape_markup(str(exc))}")
+            err.print(f"[red]Error:[/red] {_escape_markup(str(exc))}")
             raise typer.Exit(1)
 
         if json_output:
@@ -1540,7 +1555,7 @@ def workflow_add(
     # precedence over --from so a URL that would be ignored is never fetched.
     if dev:
         dev_path = Path(source).expanduser()
-        if dev_path.is_file() and dev_path.suffix in (".yml", ".yaml"):
+        if dev_path.is_file() and dev_path.suffix.lower() in (".yml", ".yaml"):
             _validate_and_install_local(dev_path, str(dev_path))
             return
         if dev_path.is_dir():
@@ -1666,8 +1681,8 @@ def workflow_add(
                 except OSError as cleanup_exc:
                     console.print(
                         "[yellow]Warning:[/yellow] Could not remove temporary "
-                        f"download file {_escape_markup(str(tmp_path))}: "
-                        f"{_escape_markup(str(cleanup_exc))}"
+                        f"workflow download file: {_escape_markup(str(cleanup_exc))} "
+                        f"(path: {_escape_markup(str(tmp_path))})"
                     )
             console.print(f"[red]Error:[/red] Failed to download workflow: {_escape_markup(str(exc))}")
             raise typer.Exit(1)
@@ -1691,15 +1706,15 @@ def workflow_add(
             except OSError as exc:
                 console.print(
                     "[yellow]Warning:[/yellow] Could not remove temporary "
-                    f"download file {_escape_markup(str(tmp_path))}: "
-                    f"{_escape_markup(str(exc))}"
+                    f"workflow download file: {_escape_markup(str(exc))} "
+                    f"(path: {_escape_markup(str(tmp_path))})"
                 )
         return
 
     # Try as a local file/directory
     source_path = Path(source)
     if source_path.exists():
-        if source_path.is_file() and source_path.suffix in (".yml", ".yaml"):
+        if source_path.is_file() and source_path.suffix.lower() in (".yml", ".yaml"):
             _validate_and_install_local(source_path, str(source_path))
             return
         elif source_path.is_dir():
@@ -2382,6 +2397,9 @@ def workflow_info(
         # Local workflow definition not found on disk; fall back to
         # catalog/registry lookup below.
         pass
+    except ValueError as exc:
+        console.print(f"[red]Error:[/red] Invalid workflow: {_escape_markup(str(exc))}")
+        raise typer.Exit(1)
 
     if definition:
         console.print(f"\n[bold cyan]{definition.name}[/bold cyan] ({definition.id})")
@@ -3148,6 +3166,102 @@ def workflow_step_catalog_remove(
     console.print(f"[green]✓[/green] Step catalog source '{removed_name}' removed")
 
 
+@workflow_overlay_app.command("add")
+def workflow_overlay_add_cmd(
+    source: Path = typer.Argument(..., help="Path to overlay YAML file"),
+    priority: int = typer.Option(
+        10,
+        "--priority",
+        help="Resolution priority (lower = higher precedence, default 10)",
+    ),
+):
+    """Add a project-local overlay for a workflow."""
+    from .overlays._commands import workflow_overlay_add
+
+    project_root = _require_specify_project()
+    if workflow_overlay_add(project_root, source, priority) is None:
+        raise typer.Exit(1)
+
+
+@workflow_overlay_app.command("set-priority")
+def workflow_overlay_set_priority_cmd(
+    workflow_id: str = typer.Argument(..., help="Workflow ID the overlay extends"),
+    overlay_id: str = typer.Argument(..., help="Overlay ID"),
+    priority: int = typer.Argument(
+        ..., help="New priority (lower = higher precedence)"
+    ),
+):
+    """Set the priority of a project-local overlay."""
+    from .overlays._commands import workflow_overlay_set_priority
+
+    project_root = _require_specify_project()
+    if not workflow_overlay_set_priority(project_root, workflow_id, overlay_id, priority):
+        raise typer.Exit(1)
+
+
+@workflow_overlay_app.command("enable")
+def workflow_overlay_enable_cmd(
+    workflow_id: str = typer.Argument(..., help="Workflow ID the overlay extends"),
+    overlay_id: str = typer.Argument(..., help="Overlay ID"),
+):
+    """Enable a project-local overlay."""
+    from .overlays._commands import workflow_overlay_enable
+
+    project_root = _require_specify_project()
+    if not workflow_overlay_enable(project_root, workflow_id, overlay_id):
+        raise typer.Exit(1)
+
+
+@workflow_overlay_app.command("disable")
+def workflow_overlay_disable_cmd(
+    workflow_id: str = typer.Argument(..., help="Workflow ID the overlay extends"),
+    overlay_id: str = typer.Argument(..., help="Overlay ID"),
+):
+    """Disable a project-local overlay."""
+    from .overlays._commands import workflow_overlay_disable
+
+    project_root = _require_specify_project()
+    if not workflow_overlay_disable(project_root, workflow_id, overlay_id):
+        raise typer.Exit(1)
+
+
+@workflow_overlay_app.command("remove")
+def workflow_overlay_remove_cmd(
+    workflow_id: str = typer.Argument(..., help="Workflow ID the overlay extends"),
+    overlay_id: str = typer.Argument(..., help="Overlay ID"),
+):
+    """Remove a project-local overlay."""
+    from .overlays._commands import workflow_overlay_remove
+
+    project_root = _require_specify_project()
+    if not workflow_overlay_remove(project_root, workflow_id, overlay_id):
+        raise typer.Exit(1)
+
+
+@workflow_overlay_app.command("list")
+def workflow_overlay_list_cmd(
+    workflow_id: str = typer.Argument(..., help="Workflow ID"),
+):
+    """List overlays for a workflow."""
+    from .overlays._commands import workflow_overlay_list
+
+    project_root = _require_specify_project()
+    if workflow_overlay_list(project_root, workflow_id) is None:
+        raise typer.Exit(1)
+
+
+@workflow_app.command("resolve")
+def workflow_resolve_cmd(
+    workflow_id: str = typer.Argument(..., help="Workflow ID to resolve"),
+):
+    """Show layer attribution for a resolved workflow."""
+    from .overlays._commands import workflow_resolve
+
+    project_root = _require_specify_project()
+    if workflow_resolve(project_root, workflow_id) is None:
+        raise typer.Exit(1)
+
+
 def register(app: typer.Typer) -> None:
     """Attach the workflow command group to the root Typer app."""
     app.add_typer(workflow_app, name="workflow")

src/specify_cli/workflows/catalog.py

@@ -364,13 +364,24 @@ class WorkflowCatalog:
             if not url:
                 continue
             self._validate_catalog_url(url)
-            try:
-                priority = int(item.get("priority", idx + 1))
-            except (TypeError, ValueError):
+            raw_priority = item.get("priority", idx + 1)
+            # bool is an int subclass: int(True) == 1 would silently accept a
+            # ``priority: true`` as priority 1. Reject it explicitly, mirroring
+            # the base CatalogStackBase loader.
+            if isinstance(raw_priority, bool):
                 raise WorkflowValidationError(
                     f"Invalid priority for catalog "
                     f"'{item.get('name', idx + 1)}': "
-                    f"expected integer, got {item.get('priority')!r}"
+                    f"expected integer, got {raw_priority!r}"
+                )
+            try:
+                priority = int(raw_priority)
+            except (TypeError, ValueError, OverflowError):
+                # OverflowError: int(float("inf")) — a ``priority: .inf``.
+                raise WorkflowValidationError(
+                    f"Invalid priority for catalog "
+                    f"'{item.get('name', idx + 1)}': "
+                    f"expected integer, got {raw_priority!r}"
                 )
             raw_install = item.get("install_allowed", False)
             if isinstance(raw_install, str):
@@ -512,8 +523,20 @@ class WorkflowCatalog:
 
         _validate_catalog_url(entry.url)
 
+        # Validate EVERY redirect hop, not just the final URL: _open_url follows
+        # redirects, so an https:// entry that 30x-redirects through http:// (or
+        # to a non-HTTPS host mid-chain) could otherwise let a network attacker
+        # rewrite the next hop and slip a payload past a final-URL-only check.
+        # redirect_validator runs before each hop; the geturl() check below is
+        # retained as a defense-in-depth backstop. Mirrors the presets/extensions
+        # catalog fix (#3523 / #3524).
+        def _validate_redirect(_old_url: str, new_url: str) -> None:
+            _validate_catalog_url(new_url)
+
         try:
-            with _open_url(entry.url, timeout=30) as resp:
+            with _open_url(
+                entry.url, timeout=30, redirect_validator=_validate_redirect
+            ) as resp:
                 _validate_catalog_url(resp.geturl())
                 data = json.loads(resp.read().decode("utf-8"))
         except Exception as exc:
@@ -685,7 +708,9 @@ class WorkflowCatalog:
         def _coerce_priority(value: Any) -> int:
             try:
                 return int(value)
-            except (TypeError, ValueError):
+            except (TypeError, ValueError, OverflowError):
+                # OverflowError: int(float("inf")) — treat an uncoercible
+                # existing priority as 0 rather than crashing 'catalog add'.
                 return 0
 
         max_priority = max(
@@ -869,7 +894,11 @@ class StepRegistry:
         import copy
         from datetime import datetime, timezone
 
-        existing = self.data["steps"].get(step_id, {})
+        raw_existing = self.data["steps"].get(step_id)
+        # Corrupted-but-parseable registries may hold non-dict entries; treat
+        # them as absent rather than crashing on existing.get() (mirrors
+        # WorkflowRegistry.add).
+        existing = raw_existing if isinstance(raw_existing, dict) else {}
         metadata_to_store = copy.deepcopy(metadata)
         metadata_to_store["installed_at"] = existing.get(
             "installed_at", datetime.now(timezone.utc).isoformat()
@@ -1007,13 +1036,23 @@ class StepCatalog:
             if not url:
                 continue
             self._validate_catalog_url(url)
-            try:
-                priority = int(item.get("priority", idx + 1))
-            except (TypeError, ValueError):
+            raw_priority = item.get("priority", idx + 1)
+            # bool is an int subclass: reject ``priority: true`` explicitly rather
+            # than silently coercing it to 1 (mirrors CatalogStackBase).
+            if isinstance(raw_priority, bool):
                 raise StepValidationError(
                     f"Invalid priority for catalog "
                     f"'{item.get('name', idx + 1)}': "
-                    f"expected integer, got {item.get('priority')!r}"
+                    f"expected integer, got {raw_priority!r}"
+                )
+            try:
+                priority = int(raw_priority)
+            except (TypeError, ValueError, OverflowError):
+                # OverflowError: int(float("inf")) — a ``priority: .inf``.
+                raise StepValidationError(
+                    f"Invalid priority for catalog "
+                    f"'{item.get('name', idx + 1)}': "
+                    f"expected integer, got {raw_priority!r}"
                 )
             raw_install = item.get("install_allowed", False)
             if isinstance(raw_install, str):
@@ -1157,8 +1196,20 @@ class StepCatalog:
 
         _validate_url(entry.url)
 
+        # Validate EVERY redirect hop, not just the final URL: _open_url follows
+        # redirects, so an https:// entry that 30x-redirects through http:// (or
+        # to a non-HTTPS host mid-chain) could otherwise let a network attacker
+        # rewrite the next hop and slip a payload past a final-URL-only check.
+        # redirect_validator runs before each hop; the geturl() check below is
+        # retained as a defense-in-depth backstop. Mirrors the presets/extensions
+        # catalog fix (#3523 / #3524).
+        def _validate_redirect(_old_url: str, new_url: str) -> None:
+            _validate_url(new_url)
+
         try:
-            with _open_url(entry.url, timeout=30) as resp:
+            with _open_url(
+                entry.url, timeout=30, redirect_validator=_validate_redirect
+            ) as resp:
                 _validate_url(resp.geturl())
                 data = json.loads(resp.read().decode("utf-8"))
         except Exception as exc:
@@ -1314,7 +1365,9 @@ class StepCatalog:
         def _coerce_priority(value: Any) -> int:
             try:
                 return int(value)
-            except (TypeError, ValueError):
+            except (TypeError, ValueError, OverflowError):
+                # OverflowError: int(float("inf")) — treat an uncoercible
+                # existing priority as 0 rather than crashing 'catalog add'.
                 return 0
 
         max_priority = max(

src/specify_cli/workflows/engine.py

@@ -79,7 +79,11 @@ class WorkflowDefinition:
     def from_yaml(cls, path: Path) -> WorkflowDefinition:
         """Load a workflow definition from a YAML file."""
         with open(path, encoding="utf-8") as f:
-            data = yaml.safe_load(f)
+            try:
+                data = yaml.safe_load(f)
+            except yaml.YAMLError as exc:
+                msg = f"Invalid YAML in {path}: {exc}"
+                raise ValueError(msg) from exc
         if not isinstance(data, dict):
             msg = f"Workflow YAML must be a mapping, got {type(data).__name__}."
             raise ValueError(msg)
@@ -88,7 +92,11 @@ class WorkflowDefinition:
     @classmethod
     def from_string(cls, content: str) -> WorkflowDefinition:
         """Load a workflow definition from a YAML string."""
-        data = yaml.safe_load(content)
+        try:
+            data = yaml.safe_load(content)
+        except yaml.YAMLError as exc:
+            msg = f"Invalid YAML: {exc}"
+            raise ValueError(msg) from exc
         if not isinstance(data, dict):
             msg = f"Workflow YAML must be a mapping, got {type(data).__name__}."
             raise ValueError(msg)
@@ -193,6 +201,20 @@ def validate_workflow(definition: WorkflowDefinition) -> list[str]:
                     f"Must be 'string', 'number', or 'boolean'."
                 )
 
+            # ``enum`` must be a list. Checked here — not only via the
+            # ``_coerce_input`` call below — because that call is reached only
+            # when a ``default`` is present, and the ``integration: auto`` case
+            # strips ``enum`` before coercing; a scalar/string ``enum`` on an
+            # input with no default (or the auto-integration default) would
+            # otherwise slip through here and then crash ``_resolve_inputs`` with
+            # a raw ``TypeError`` at run time. ``None`` means "no enum".
+            enum_values = input_def.get("enum")
+            if enum_values is not None and not isinstance(enum_values, list):
+                errors.append(
+                    f"Input {input_name!r} has invalid 'enum': must be a list, "
+                    f"got {type(enum_values).__name__}."
+                )
+
             # Validate the default eagerly so authoring mistakes (e.g. a
             # default not in the declared enum, or a non-numeric default for
             # a number input) surface at install/validation time instead of
@@ -201,13 +223,28 @@ def validate_workflow(definition: WorkflowDefinition) -> list[str]:
             # enum-membership check is exempted for that exact case — the
             # declared type is still enforced (e.g. ``type: number`` paired
             # with ``default: "auto"`` is still rejected).
+            enum_is_valid = enum_values is None or isinstance(enum_values, list)
             if "default" in input_def:
                 default_value = input_def["default"]
                 is_auto_integration = (
                     input_name == "integration" and default_value == "auto"
                 )
+                # Strip ``enum`` from the definition handed to ``_coerce_input``
+                # when either:
+                #   * this is the auto-integration sentinel (enum-membership is
+                #     a runtime concern, exempted for ``"auto"``), or
+                #   * the ``enum`` is malformed (non-list) and already reported
+                #     above — leaving it in would make ``_coerce_input`` re-raise
+                #     the same enum-shape error re-framed as an "invalid default"
+                #     (a confusing duplicate).
+                # Removing *only* ``enum`` (rather than skipping the check
+                # entirely) preserves the default's type validation: a
+                # ``type: string`` input with ``default: 5, enum: 5`` still
+                # reports the wrong-typed default alongside the enum error,
+                # instead of hiding it.
+                strip_enum = is_auto_integration or not enum_is_valid
                 validation_input_def: dict[str, Any] = input_def
-                if is_auto_integration and "enum" in input_def:
+                if strip_enum and "enum" in input_def:
                     validation_input_def = {
                         key: value
                         for key, value in input_def.items()
@@ -727,13 +764,24 @@ class WorkflowEngine:
         ValueError:
             If the workflow YAML is invalid.
         """
+        from .overlays import WorkflowResolver
+
         path = Path(source).expanduser()
 
         # Try as a direct file path first
         if path.suffix.lower() in (".yml", ".yaml") and path.is_file():
             return WorkflowDefinition.from_yaml(path)
 
-        # Try as an installed workflow ID
+        # Try as an installed workflow ID, resolving any overlays.
+        resolver = WorkflowResolver(self.project_root)
+        try:
+            return resolver.resolve(str(source))
+        except FileNotFoundError:
+            # Fall back to the direct workflow.yml path so callers still get
+            # the original error when the workflow id is not installed.
+            pass
+
+        # Legacy direct path check for workflows installed without registry entries.
         installed_path = (
             self.project_root
             / ".specify"
@@ -1381,11 +1429,18 @@ class WorkflowEngine:
             # definition (``string`` rejects non-strings, ``number`` rejects
             # bools and uncoercible values, ``boolean`` rejects non-bools),
             # so ill-typed values still fail fast here.
+            #
+            # ``execute()`` accepts unvalidated definitions, so a malformed
+            # (non-list) ``enum`` can reach here. Only strip a *list* ``enum``:
+            # a scalar/string ``enum`` must stay in the definition so
+            # ``_coerce_input`` raises the clean shape ``ValueError`` instead of
+            # being silently exempted by the ``auto`` membership skip (which
+            # would otherwise let ``enum: 5`` resolve successfully).
             coerce_input_def = input_def
             if (
                 name == "integration"
                 and value == "auto"
-                and "enum" in input_def
+                and isinstance(input_def.get("enum"), list)
             ):
                 coerce_input_def = {
                     key: val
@@ -1431,6 +1486,22 @@ class WorkflowEngine:
         input_type = input_def.get("type", "string")
         enum_values = input_def.get("enum")
 
+        # ``enum`` must be a list. A scalar (``enum: 5``, ``enum: true``) makes
+        # the ``value not in enum_values`` membership test below raise a raw
+        # ``TypeError`` ("argument of type 'int' is not ... iterable"), which
+        # escapes ``validate_workflow``'s ``except ValueError`` and breaks its
+        # "return errors, never raise" contract — and crashes ``_resolve_inputs``
+        # outright at run time. A bare string is just as wrong: ``value in "abc"``
+        # is a silent substring/character test, not enum membership. Require a
+        # list so both forms fail fast with a clear message. ``None`` means "no
+        # enum" and is left alone.
+        if enum_values is not None and not isinstance(enum_values, list):
+            msg = (
+                f"Input {name!r} has invalid 'enum': must be a list, got "
+                f"{type(enum_values).__name__}."
+            )
+            raise ValueError(msg)
+
         if input_type == "number":
             # Reject bools explicitly: ``bool`` is a subclass of ``int`` so
             # ``float(True)`` succeeds and would silently coerce a YAML

src/specify_cli/workflows/expressions.py

@@ -535,6 +535,10 @@ def _evaluate_simple_expression(expr: str, namespace: dict[str, Any]) -> Any:
         items = [
             _evaluate_simple_expression(i.strip(), namespace)
             for i in _split_top_level_commas(inner)
+            # Drop empty segments from trailing/leading/double commas ([1, 2,] ->
+            # [1, 2], not [1, 2, None]). An intentional empty-string element
+            # ('') strips to "''" (truthy), so ['', 'a'] is preserved.
+            if i.strip()
         ]
         return items
 

src/specify_cli/workflows/overlays/__init__.py

@@ -0,0 +1,95 @@
+"""Workflow overlay resolver — composes installed workflows from layers."""
+
+from __future__ import annotations
+
+from pathlib import Path
+
+from ..engine import WorkflowDefinition
+from .composer import StepListComposer
+from .layer_sources import (
+    BaseWorkflowSource,
+    Layer,
+    ProjectOverlaySource,
+)
+from .merge import ComposedStep
+from .schema import _RESERVED_WORKFLOW_IDS, _SAFE_ID_PATTERN
+
+
+def _validate_workflow_id(workflow_id: str) -> None:
+    """Reject workflow IDs that are unsafe as installed-storage path segments."""
+    if (
+        not isinstance(workflow_id, str)
+        or not _SAFE_ID_PATTERN.fullmatch(workflow_id)
+        or workflow_id in _RESERVED_WORKFLOW_IDS
+    ):
+        raise ValueError(f"Invalid workflow ID: {workflow_id!r}")
+
+
+class WorkflowResolver:
+    """Resolves a workflow ID to its composed ``WorkflowDefinition``.
+
+    Collects layers from two tiers:
+    - project-local overlays (``.specify/workflows/overlays/<id>/*.yml``)
+    - the base workflow itself (``.specify/workflows/<id>/workflow.yml``)
+
+    Resolution is lower-wins: overlays with lower priority numbers are applied
+    later and override earlier edits on the same anchors.
+    """
+
+    def __init__(self, project_root: Path) -> None:
+        self.project_root = project_root
+        self._sources = [
+            ProjectOverlaySource(project_root),
+            BaseWorkflowSource(project_root),
+        ]
+        self._composer = StepListComposer()
+
+    def collect_all_layers(
+        self, workflow_id: str, *, include_disabled: bool = False
+    ) -> list[Layer]:
+        """Collect overlays sorted by precedence, followed by the base layer.
+
+        Lower priority numbers win. Ties are sorted alphabetically by source,
+        matching ``PresetRegistry.list_by_priority()``. The base workflow is a
+        foundation rather than a precedence candidate, so it is kept separate.
+        """
+        _validate_workflow_id(workflow_id)
+
+        all_layers: list[Layer] = []
+        for source in self._sources:
+            all_layers.extend(
+                source.collect(workflow_id, include_disabled=include_disabled)
+            )
+
+        overlays = [layer for layer in all_layers if layer.tier != "base"]
+        base_layers = [layer for layer in all_layers if layer.tier == "base"]
+        return (
+            sorted(overlays, key=lambda layer: (layer.priority, layer.source))
+            + base_layers
+        )
+
+    def resolve(self, workflow_id: str) -> WorkflowDefinition:
+        """Resolve a workflow ID to its composed definition.
+
+        This method composes layers but does not validate workflow semantics;
+        callers should validate the returned definition when needed.
+
+        Raises:
+            FileNotFoundError: if the workflow cannot be found.
+            ValueError: if layer collection/composition fails.
+        """
+        layers = self.collect_all_layers(workflow_id)
+        definition, _ = self._composer.compose(layers)
+        if definition is None:
+            raise FileNotFoundError(f"Workflow not found: {workflow_id}")
+        return definition
+
+    def resolve_with_layers(
+        self, workflow_id: str
+    ) -> tuple[WorkflowDefinition, list[Layer], list[ComposedStep]]:
+        """Resolve a workflow and return its definition plus layer attribution."""
+        layers = self.collect_all_layers(workflow_id)
+        definition, attribution = self._composer.compose(layers)
+        if definition is None:
+            raise FileNotFoundError(f"Workflow not found: {workflow_id}")
+        return definition, layers, attribution

src/specify_cli/workflows/overlays/_commands.py

@@ -0,0 +1,442 @@
+"""CLI handlers for ``specify workflow overlay *`` and ``specify workflow resolve``."""
+
+from __future__ import annotations
+
+from pathlib import Path
+from typing import Any
+
+import typer
+import yaml
+
+from ..._console import console, err_console
+from ...extensions import normalize_priority
+from .._commands import (
+    _commit_workflow_file,
+    _discard_committed_backup_file,
+    _reject_unsafe_dir,
+    _reject_unsafe_workflow_storage,
+    _safe_discard_staged_workflow_file,
+    _stage_workflow_file,
+)
+from . import WorkflowResolver
+from .schema import _RESERVED_WORKFLOW_IDS, _SAFE_ID_PATTERN, validate_overlay_yaml
+
+
+def _validate_overlay_id_or_exit(id_value: str, label: str) -> None:
+    """Validate a single-segment overlay/workflow id from CLI arguments."""
+    if not isinstance(id_value, str) or not id_value:
+        err_console.print(f"[red]Error:[/red] {label} is required and must be a non-empty string.")
+        raise typer.Exit(1)
+    if not _SAFE_ID_PATTERN.fullmatch(id_value):
+        err_console.print(
+            f"[red]Error:[/red] Invalid {label} {id_value!r}: "
+            "only lowercase letters, digits, and hyphens are allowed."
+        )
+        raise typer.Exit(1)
+
+
+def _validate_workflow_id_or_exit(workflow_id: str) -> None:
+    """Validate a workflow id, treating the overlay root as reserved."""
+    _validate_overlay_id_or_exit(workflow_id, "workflow ID")
+    if workflow_id in _RESERVED_WORKFLOW_IDS:
+        err_console.print(
+            f"[red]Error:[/red] Invalid workflow ID {workflow_id!r}: "
+            "reserved name."
+        )
+        raise typer.Exit(1)
+
+
+def _overlay_root(project_root: Path) -> Path:
+    """Return the project-local overlay root after rejecting unsafe ancestors."""
+    _reject_unsafe_workflow_storage(project_root)
+    root = project_root / ".specify" / "workflows" / "overlays"
+    _reject_unsafe_dir(root, ".specify/workflows/overlays")
+    return root
+
+
+def _project_overlay_dir(project_root: Path, workflow_id: str) -> Path:
+    """Return the project-local overlay directory for a workflow id.
+
+    Raises typer.Exit if the resolved path escapes the overlay root.
+    """
+    _validate_workflow_id_or_exit(workflow_id)
+    root = _overlay_root(project_root)
+    target = root / workflow_id
+    return _ensure_contained_dir(target, root)
+
+
+def _ensure_contained_dir(path: Path, root: Path) -> Path:
+    """Ensure *path* resolves inside *root* and is not a symlink.
+
+    Returns *path* if safe. Raises typer.Exit on traversal or symlink.
+    """
+    _reject_unsafe_dir(root, ".specify/workflows/overlays")
+    if path.is_symlink():
+        err_console.print(
+            f"[red]Error:[/red] Refusing to use symlinked path {path}."
+        )
+        raise typer.Exit(1)
+    if path.exists() and not path.is_dir():
+        err_console.print(
+            f"[red]Error:[/red] Overlay directory path is not a directory: {path}."
+        )
+        raise typer.Exit(1)
+    try:
+        resolved = path.resolve()
+        root_resolved = root.resolve()
+        resolved.relative_to(root_resolved)
+    except ValueError:
+        err_console.print(
+            f"[red]Error:[/red] Path traversal detected: {path} is outside the allowed directory."
+        )
+        raise typer.Exit(1)
+    return path
+
+
+def _find_overlay_file(project_root: Path, workflow_id: str, overlay_id: str) -> Path | None:
+    """Locate a project-local overlay file by its manifest ID, not filename.
+
+    Scans all YAML files in the overlay directory and matches on the ``id``
+    field inside each manifest. This aligns with ``ProjectOverlaySource.collect()``
+    which also derives identity from the manifest, not the filename.
+    """
+    _validate_workflow_id_or_exit(workflow_id)
+    _validate_overlay_id_or_exit(overlay_id, "overlay ID")
+    overlay_dir = _project_overlay_dir(project_root, workflow_id)
+    if not overlay_dir.is_dir():
+        return None
+    try:
+        entries = sorted(overlay_dir.iterdir())
+    except OSError:
+        return None
+    matches: list[Path] = []
+    for path in entries:
+        if not path.is_file() or path.suffix not in (".yml", ".yaml"):
+            continue
+        if path.is_symlink():
+            continue
+        data, _ = _read_overlay(path)
+        if data is None:
+            continue
+        if data.get("id") == overlay_id:
+            matches.append(path)
+    if len(matches) > 1:
+        paths = ", ".join(str(path) for path in matches)
+        err_console.print(
+            f"[red]Error:[/red] Duplicate overlay ID '{overlay_id}' in {paths}. "
+            "Resolve the duplicate manifest IDs before continuing."
+        )
+        raise typer.Exit(1)
+    return matches[0] if matches else None
+
+
+def _ensure_contained_path(path: Path, root: Path) -> Path:
+    """Return *path* only if it resolves inside *root*; otherwise raise typer.Exit."""
+    _reject_unsafe_dir(root, ".specify/workflows/overlays")
+    if path.is_symlink():
+        err_console.print(
+            f"[red]Error:[/red] Refusing to use symlinked path {path}."
+        )
+        raise typer.Exit(1)
+    try:
+        resolved = path.resolve()
+        root_resolved = root.resolve()
+        resolved.relative_to(root_resolved)
+    except ValueError:
+        err_console.print(
+            f"[red]Error:[/red] Path traversal detected: {path} is outside the allowed directory."
+        )
+        raise typer.Exit(1)
+    return path
+
+
+def _read_overlay(path: Path) -> tuple[dict[str, Any] | None, list[str]]:
+    """Read and parse an overlay YAML file, returning (data, errors)."""
+    try:
+        content = path.read_text(encoding="utf-8")
+    except (OSError, UnicodeDecodeError) as exc:
+        return None, [f"Failed to read {path}: {exc}"]
+    try:
+        data = yaml.safe_load(content)
+    except yaml.YAMLError as exc:
+        return None, [f"Invalid YAML in {path}: {exc}"]
+    if not isinstance(data, dict):
+        return None, [f"Overlay {path} must be a YAML mapping."]
+    return data, []
+
+
+def workflow_overlay_add(
+    project_root: Path,
+    source: Path,
+    priority: int | None = None,
+) -> Path | None:
+    """Add a project-local overlay from a YAML file.
+
+    Returns the path of the installed overlay file, or None on failure.
+    """
+    _reject_unsafe_workflow_storage(project_root)
+    data, errors = _read_overlay(source)
+    if data is None:
+        for err in errors:
+            err_console.print(f"[red]Error:[/red] {err}")
+        return None
+
+    # Apply --priority override before validation so a valid CLI priority
+    # can fix a missing or invalid priority in the file.
+    if priority is not None:
+        if isinstance(priority, bool) or not isinstance(priority, int) or priority < 1:
+            err_console.print("[red]Error:[/red] Priority must be >= 1.")
+            return None
+        data["priority"] = normalize_priority(priority)
+
+    overlay, validation_errors = validate_overlay_yaml(data)
+    if overlay is None:
+        err_console.print("[red]Error:[/red] Overlay validation failed:")
+        for err in validation_errors:
+            err_console.print(f"  \u2022 {err}")
+        return None
+    data["priority"] = overlay.priority
+
+    target_dir = _project_overlay_dir(project_root, overlay.extends)
+    # Reuse an existing .yaml file so we don't create a duplicate .yml layer.
+    existing = _find_overlay_file(project_root, overlay.extends, overlay.id)
+    if existing is not None:
+        target_path = existing
+    else:
+        target_path = _ensure_contained_path(
+            target_dir / f"{overlay.id}.yml", _overlay_root(project_root)
+        )
+
+    backup: Path | None = None
+    try:
+        target_dir.mkdir(parents=True, exist_ok=True)
+        existed_before = target_path.exists()
+        staged = _stage_workflow_file(target_path.parent)
+        try:
+            staged.write_bytes(yaml.safe_dump(data, sort_keys=False).encode("utf-8"))
+            backup = _commit_workflow_file(staged, target_path, existed_before)
+        except BaseException:
+            _safe_discard_staged_workflow_file(
+                staged, target_path.parent, existed_before
+            )
+            raise
+    except OSError as exc:
+        err_console.print(f"[red]Error:[/red] Failed to write overlay: {exc}")
+        return None
+    _discard_committed_backup_file(backup)
+
+    console.print(
+        f"[green]\u2713[/green] Overlay '{overlay.id}' added for workflow '{overlay.extends}'"
+    )
+    return target_path
+
+
+def _update_overlay_field(
+    project_root: Path,
+    workflow_id: str,
+    overlay_id: str,
+    field: str,
+    value: Any,
+) -> bool:
+    """Update a single field in a project-local overlay file."""
+    _reject_unsafe_workflow_storage(project_root)
+    path = _find_overlay_file(project_root, workflow_id, overlay_id)
+    if path is None:
+        err_console.print(
+            f"[red]Error:[/red] Overlay '{overlay_id}' not found for workflow '{workflow_id}'"
+        )
+        return False
+
+    data, errors = _read_overlay(path)
+    if data is None:
+        for err in errors:
+            err_console.print(f"[red]Error:[/red] {err}")
+        return False
+
+    data[field] = value
+    overlay, validation_errors = validate_overlay_yaml(data)
+    if overlay is None:
+        err_console.print("[red]Error:[/red] Overlay validation failed:")
+        for err in validation_errors:
+            err_console.print(f"  \u2022 {err}")
+        return False
+
+    backup: Path | None = None
+    try:
+        existed_before = path.exists()
+        staged = _stage_workflow_file(path.parent)
+        try:
+            staged.write_bytes(yaml.safe_dump(data, sort_keys=False).encode("utf-8"))
+            backup = _commit_workflow_file(staged, path, existed_before)
+        except BaseException:
+            _safe_discard_staged_workflow_file(staged, path.parent, existed_before)
+            raise
+    except OSError as exc:
+        err_console.print(f"[red]Error:[/red] Failed to write overlay: {exc}")
+        return False
+    _discard_committed_backup_file(backup)
+
+    return True
+
+
+def workflow_overlay_set_priority(
+    project_root: Path,
+    workflow_id: str,
+    overlay_id: str,
+    priority: int,
+) -> bool:
+    """Set the priority of a project-local overlay."""
+    if isinstance(priority, bool) or not isinstance(priority, int) or priority < 1:
+        err_console.print("[red]Error:[/red] Priority must be >= 1.")
+        raise typer.Exit(1)
+    normalized_priority = normalize_priority(priority)
+    if _update_overlay_field(
+        project_root, workflow_id, overlay_id, "priority", normalized_priority
+    ):
+        console.print(
+            f"[green]\u2713[/green] Priority of overlay '{overlay_id}' set to {normalized_priority}"
+        )
+        return True
+    return False
+
+
+def workflow_overlay_enable(
+    project_root: Path,
+    workflow_id: str,
+    overlay_id: str,
+) -> bool:
+    """Enable a project-local overlay."""
+    if _update_overlay_field(project_root, workflow_id, overlay_id, "enabled", True):
+        console.print(f"[green]\u2713[/green] Overlay '{overlay_id}' enabled")
+        return True
+    return False
+
+
+def workflow_overlay_disable(
+    project_root: Path,
+    workflow_id: str,
+    overlay_id: str,
+) -> bool:
+    """Disable a project-local overlay."""
+    if _update_overlay_field(project_root, workflow_id, overlay_id, "enabled", False):
+        console.print(f"[green]\u2713[/green] Overlay '{overlay_id}' disabled")
+        return True
+    return False
+
+
+def workflow_overlay_remove(
+    project_root: Path,
+    workflow_id: str,
+    overlay_id: str,
+) -> bool:
+    """Remove a project-local overlay file."""
+    _reject_unsafe_workflow_storage(project_root)
+    path = _find_overlay_file(project_root, workflow_id, overlay_id)
+    if path is None:
+        err_console.print(
+            f"[red]Error:[/red] Overlay '{overlay_id}' not found for workflow '{workflow_id}'"
+        )
+        return False
+
+    try:
+        path.unlink()
+    except OSError as exc:
+        err_console.print(f"[red]Error:[/red] Failed to remove overlay: {exc}")
+        return False
+
+    console.print(f"[green]\u2713[/green] Overlay '{overlay_id}' removed")
+    return True
+
+
+def workflow_overlay_list(project_root: Path, workflow_id: str) -> list[dict[str, Any]] | None:
+    """List all overlays for a workflow and print a summary table.
+
+    Returns the raw list data for machine-readable callers, or None on error.
+    """
+    _reject_unsafe_workflow_storage(project_root)
+    _validate_workflow_id_or_exit(workflow_id)
+    resolver = WorkflowResolver(project_root)
+    try:
+        layers = resolver.collect_all_layers(workflow_id, include_disabled=True)
+    except ValueError as exc:
+        err_console.print(f"[red]Error:[/red] {exc}")
+        return None
+    overlays = [layer for layer in layers if layer.tier != "base"]
+
+    if not overlays:
+        console.print(f"[yellow]No overlays found for workflow '{workflow_id}'.[/yellow]")
+        return []
+
+    console.print(f"Overlays for workflow '{workflow_id}':")
+    rows: list[dict[str, Any]] = []
+    for layer in overlays:
+        overlay = layer.content
+        rows.append({
+            "id": overlay.id,
+            "source": layer.source,
+            "tier": layer.tier,
+            "priority": normalize_priority(overlay.priority),
+            "enabled": overlay.enabled,
+            "path": str(layer.path) if layer.path else None,
+        })
+        enabled_marker = "enabled" if overlay.enabled else "disabled"
+        console.print(
+            f"  \u2022 {overlay.id} (priority={normalize_priority(overlay.priority)}, "
+            f"source={layer.source}, {enabled_marker})"
+        )
+    return rows
+
+
+def workflow_resolve(project_root: Path, workflow_id: str) -> dict[str, Any] | None:
+    """Print layer attribution for a resolved workflow.
+
+    Returns a serializable attribution payload.
+    """
+    _reject_unsafe_workflow_storage(project_root)
+    _validate_workflow_id_or_exit(workflow_id)
+    resolver = WorkflowResolver(project_root)
+    try:
+        definition, layers, attribution = resolver.resolve_with_layers(workflow_id)
+    except FileNotFoundError:
+        err_console.print(
+            f"[red]Error:[/red] Workflow '{workflow_id}' not found"
+        )
+        return None
+    except ValueError as exc:
+        err_console.print(f"[red]Error:[/red] {exc}")
+        return None
+
+    console.print(f"Resolved workflow '{workflow_id}':")
+    console.print("Layers (highest precedence first):")
+    for layer in layers:
+        priority = (
+            "n/a" if layer.tier == "base" else str(normalize_priority(layer.priority))
+        )
+        console.print(
+            f"  \u2022 [{layer.tier}] {layer.source} "
+            f"(priority={priority})"
+        )
+
+    console.print("Step attribution:")
+    for composed in attribution:
+        console.print(f"  \u2022 {composed.step_id}: {composed.source}")
+
+    return {
+        "workflow_id": workflow_id,
+        "layers": [
+            {
+                "source": layer.source,
+                "tier": layer.tier,
+                "priority": (
+                    None
+                    if layer.tier == "base"
+                    else normalize_priority(layer.priority)
+                ),
+            }
+            for layer in layers
+        ],
+        "attribution": [
+            {"step_id": composed.step_id, "source": composed.source}
+            for composed in attribution
+        ],
+    }

src/specify_cli/workflows/overlays/composer.py

@@ -0,0 +1,97 @@
+"""Workflow overlay composer — builds a WorkflowDefinition from layers."""
+
+from __future__ import annotations
+
+from typing import Any
+
+from ..engine import WorkflowDefinition
+from .layer_sources import Layer
+from .merge import OverlayLayer, merge_steps, validate_edits
+
+
+class StepListComposer:
+    """Compose a workflow from a base layer and overlay layers.
+
+    - The base layer (tier="base") provides the full step list.
+    - Overlay layers provide edit operations.
+    - Overlays are applied in merge order: highest priority number first,
+      lowest last, so lower priority numbers win. Ties are applied by overlay
+      ID, with the alphabetically last ID winning.
+    - Returns a parsed WorkflowDefinition; callers must validate separately.
+    """
+
+    def compose(
+        self, layers: list[Layer]
+    ) -> tuple[WorkflowDefinition | None, list]:
+        """Compose a ``WorkflowDefinition`` from the given layers.
+
+        Returns ``(None, [])`` when no base layer is present.
+        """
+        base_layer: Layer | None = None
+        overlay_layers: list[Layer] = []
+        for layer in layers:
+            if layer.tier == "base":
+                base_layer = layer
+            else:
+                overlay_layers.append(layer)
+
+        if base_layer is None or base_layer.path is None:
+            return None, []
+
+        # Read the base workflow definition from disk.
+        base_definition = WorkflowDefinition.from_yaml(base_layer.path)
+        base_steps = base_definition.data.get("steps", [])
+        if not isinstance(base_steps, list):
+            # Preserve the invalid definition intact so validate_workflow can
+            # report "'steps' must be a list." to the caller; coercing to []
+            # here would mask that error.
+            return base_definition, []
+
+        # Last applied wins, so apply lower priority numbers last.
+        merge_order = sorted(
+            overlay_layers,
+            key=lambda layer: (-layer.priority, layer.content.id),
+        )
+
+        # Validate edits against base anchors before mutation.
+        base_step_ids = self._collect_base_step_ids(base_steps)
+        for layer in merge_order:
+            edit_errors = validate_edits(layer.content.edits, base_step_ids)
+            if edit_errors:
+                raise ValueError(
+                    f"Overlay '{layer.content.id}' has invalid edits:\n  - "
+                    + "\n  - ".join(edit_errors)
+                )
+
+        composed_steps, attribution = merge_steps(
+            base_steps,
+            [OverlayLayer(layer.content, layer.source) for layer in merge_order],
+        )
+
+        # Build composed data while preserving all non-step fields from base.
+        composed_data: dict[str, Any] = dict(base_definition.data)
+        composed_data["steps"] = composed_steps
+
+        composed_definition = WorkflowDefinition(composed_data, source_path=base_layer.path)
+
+        return composed_definition, attribution
+
+    def _collect_base_step_ids(self, steps: list[dict[str, Any]]) -> set[str]:
+        """Collect all base step IDs reachable in the step tree."""
+        ids: set[str] = set()
+        for step in steps:
+            if not isinstance(step, dict):
+                continue
+            step_id = step.get("id")
+            if isinstance(step_id, str):
+                ids.add(step_id)
+            for key in ("then", "else", "steps", "default"):
+                nested = step.get(key)
+                if isinstance(nested, list):
+                    ids.update(self._collect_base_step_ids(nested))
+            cases = step.get("cases")
+            if isinstance(cases, dict):
+                for case_steps in cases.values():
+                    if isinstance(case_steps, list):
+                        ids.update(self._collect_base_step_ids(case_steps))
+        return ids

src/specify_cli/workflows/overlays/layer_sources.py

@@ -0,0 +1,234 @@
+"""Workflow overlay layer sources."""
+
+from __future__ import annotations
+
+from dataclasses import dataclass
+from pathlib import Path
+
+import yaml
+
+from .schema import Overlay, _RESERVED_WORKFLOW_IDS, _SAFE_ID_PATTERN, validate_overlay_yaml
+
+
+@dataclass
+class Layer:
+    """A single layer in the workflow overlay stack."""
+
+    content: Overlay
+    source: str
+    tier: str
+    priority: int
+    path: Path | None = None
+
+
+class OverlayLoadError(ValueError):
+    """Raised when an overlay file cannot be loaded or validated."""
+
+    def __init__(self, path: Path, errors: list[str]) -> None:
+        self.path = path
+        self.errors = errors
+        super().__init__(f"Invalid overlay {path}:\n  - " + "\n  - ".join(errors))
+
+
+def _validate_workflow_id(workflow_id: str, context_path: Path) -> None:
+    """Raise OverlayLoadError if workflow_id is not a safe path-segment identifier.
+
+    Mirrors the same check performed by WorkflowResolver so layer sources are
+    safe to call directly, without going through the resolver.
+    """
+    if (
+        not isinstance(workflow_id, str)
+        or not _SAFE_ID_PATTERN.fullmatch(workflow_id)
+        or workflow_id in _RESERVED_WORKFLOW_IDS
+    ):
+        raise OverlayLoadError(
+            context_path,
+            [f"Invalid workflow ID: {workflow_id!r}"],
+        )
+
+
+def _ensure_contained_dir(path: Path, root: Path) -> None:
+    """Raise OverlayLoadError if *path* is a symlink, a non-directory, or escapes *root*.
+
+    Mirrors the logic of ``_ensure_contained_dir`` in ``overlays/_commands.py``
+    but raises ``OverlayLoadError`` instead of ``typer.Exit`` so layer sources
+    can enforce the same invariants without a CLI dependency.
+
+    The caller is responsible for ensuring *root* itself is already validated
+    (e.g. via ``_resolve_project_overlay_root``).
+    """
+    if path.is_symlink():
+        raise OverlayLoadError(path, ["Symlinked overlay directories are not allowed"])
+    if path.exists() and not path.is_dir():
+        raise OverlayLoadError(path, ["Overlay directory path is not a directory"])
+    try:
+        path.resolve().relative_to(root.resolve())
+    except ValueError:
+        raise OverlayLoadError(
+            path, ["Path traversal detected: directory escapes allowed root"]
+        ) from None
+
+
+def _resolve_workflows_root(project_root: Path) -> Path:
+    """Return the workflow storage root after rejecting unsafe ancestors."""
+    project_root_resolved = project_root.resolve()
+    workflows_root = project_root / ".specify" / "workflows"
+
+    current = project_root
+    for part in (".specify", "workflows"):
+        current = current / part
+        if current.is_symlink():
+            raise OverlayLoadError(
+                current,
+                [f"Symlinked workflow directories are not allowed ({current})"],
+            )
+        if current.exists() and not current.is_dir():
+            raise OverlayLoadError(
+                current,
+                [f"Workflow directory path is not a directory ({current})"],
+            )
+
+    try:
+        workflows_root.resolve().relative_to(project_root_resolved)
+    except ValueError:
+        raise OverlayLoadError(
+            workflows_root,
+            ["Workflow directory escapes the project root"],
+        ) from None
+    return workflows_root
+
+
+def _resolve_project_overlay_root(project_root: Path) -> Path:
+    """Return the unresolved overlay root after rejecting unsafe ancestors."""
+    workflows_root = _resolve_workflows_root(project_root)
+    overlays_root = workflows_root / "overlays"
+    if overlays_root.is_symlink():
+        raise OverlayLoadError(
+            overlays_root,
+            [f"Symlinked overlay directories are not allowed ({overlays_root})"],
+        )
+    if overlays_root.exists() and not overlays_root.is_dir():
+        raise OverlayLoadError(
+            overlays_root,
+            [f"Overlay directory path is not a directory ({overlays_root})"],
+        )
+    return overlays_root
+
+
+class ProjectOverlaySource:
+    """Project-local overlays: ``.specify/workflows/overlays/<id>/*.yml``."""
+
+    tier = "project-overlay"
+
+    def __init__(self, project_root: Path) -> None:
+        self.project_root = project_root
+        self.overlays_dir = project_root / ".specify" / "workflows" / "overlays"
+
+    def collect(self, workflow_id: str, *, include_disabled: bool = False) -> list[Layer]:
+        """Collect project-local overlays for the given workflow id.
+
+        Args:
+            workflow_id: Workflow identifier whose overlay directory to scan.
+            include_disabled: When True, return disabled overlays for
+                management/list views. Resolution paths keep the default False.
+        """
+        self.overlays_dir = _resolve_project_overlay_root(self.project_root)
+        _validate_workflow_id(workflow_id, self.overlays_dir)
+        workflow_overlay_dir = self.overlays_dir / workflow_id
+        _ensure_contained_dir(workflow_overlay_dir, self.overlays_dir)
+        if not workflow_overlay_dir.is_dir():
+            return []
+        layers: list[Layer] = []
+        overlay_paths_by_id: dict[str, Path] = {}
+        try:
+            entries = sorted(workflow_overlay_dir.iterdir())
+        except OSError as exc:
+            raise OverlayLoadError(
+                workflow_overlay_dir, [f"Cannot enumerate overlays: {exc}"]
+            ) from exc
+        for path in entries:
+            if not path.is_file() or path.suffix not in (".yml", ".yaml"):
+                continue
+            if path.is_symlink():
+                raise OverlayLoadError(path, ["Symlinked overlay files are not allowed"])
+            try:
+                data = yaml.safe_load(path.read_text(encoding="utf-8")) or {}
+            except yaml.YAMLError as exc:
+                raise OverlayLoadError(path, [f"Invalid YAML: {exc}"]) from exc
+            except (OSError, UnicodeDecodeError) as exc:
+                raise OverlayLoadError(path, [f"Cannot load overlay: {exc}"]) from exc
+            if (
+                not include_disabled
+                and isinstance(data, dict)
+                and data.get("enabled", True) is False
+            ):
+                continue
+            overlay, errors = validate_overlay_yaml(data)
+            if overlay is None or errors:
+                raise OverlayLoadError(path, errors)
+            if overlay.extends != workflow_id:
+                raise OverlayLoadError(
+                    path,
+                    [
+                        f"Overlay extends {overlay.extends!r}, but is stored under "
+                        f"workflow {workflow_id!r}."
+                    ],
+                )
+            first_path = overlay_paths_by_id.get(overlay.id)
+            if first_path is not None:
+                raise OverlayLoadError(
+                    path,
+                    [
+                        f"Duplicate overlay id {overlay.id!r}; also declared in "
+                        f"{first_path}."
+                    ],
+                )
+            overlay_paths_by_id[overlay.id] = path
+            layers.append(
+                Layer(
+                    content=overlay,
+                    source=f"project:{overlay.id}",
+                    tier=self.tier,
+                    priority=overlay.priority,
+                    path=path,
+                )
+            )
+        return layers
+
+
+class BaseWorkflowSource:
+    """Base workflow layer: ``.specify/workflows/<id>/workflow.yml``."""
+
+    tier = "base"
+
+    def __init__(self, project_root: Path) -> None:
+        self.project_root = project_root
+        self.workflows_dir = project_root / ".specify" / "workflows"
+
+    def collect(self, workflow_id: str, *, include_disabled: bool = False) -> list[Layer]:
+        """Return the base workflow as a single layer if it exists."""
+        self.workflows_dir = _resolve_workflows_root(self.project_root)
+        _validate_workflow_id(workflow_id, self.workflows_dir)
+        workflow_dir = self.workflows_dir / workflow_id
+        _ensure_contained_dir(workflow_dir, self.workflows_dir)
+        path = workflow_dir / "workflow.yml"
+        if path.is_symlink():
+            raise OverlayLoadError(path, ["Symlinked workflow files are not allowed"])
+        if not path.is_file():
+            return []
+        # The base layer is represented by an Overlay with empty edits.
+        overlay = Overlay(
+            id=workflow_id,
+            extends=workflow_id,
+            priority=0,
+            edits=[],
+        )
+        return [
+            Layer(
+                content=overlay,
+                source="base",
+                tier=self.tier,
+                priority=0,
+                path=path,
+            )
+        ]

src/specify_cli/workflows/overlays/merge.py

@@ -0,0 +1,395 @@
+"""Pure-function merge engine for workflow step lists."""
+
+from __future__ import annotations
+
+import copy
+from dataclasses import dataclass
+from typing import Any
+
+from .schema import VALID_OPERATIONS, Overlay, OverlayEdit
+
+
+@dataclass(frozen=True)
+class ComposedStep:
+    """Attribution tracking for a single composed step."""
+
+    step_id: str
+    source: str
+
+
+@dataclass(frozen=True)
+class OverlayLayer:
+    """An overlay together with its layer source for attribution."""
+
+    overlay: Overlay
+    source: str
+
+
+# Nested step keys that may contain a list of steps.
+_NESTED_LIST_KEYS = ("then", "else", "steps", "default")
+
+
+def find_step(
+    steps: list[dict[str, Any]], step_id: str
+) -> tuple[list[dict[str, Any]], int] | None:
+    """Recursively locate a step by ID and return its (parent_list, index).
+
+    Searches flat lists and nested lists inside ``then``, ``else``, ``steps``,
+    ``default``, and ``cases.*``.  Does *not* descend into ``fan-out`` template
+    steps because those are runtime-multiplied stamps, not uniquely-addressable
+    nodes.
+    """
+    for i, step in enumerate(steps):
+        if not isinstance(step, dict):
+            continue
+        if step.get("id") == step_id:
+            return (steps, i)
+        for key in _NESTED_LIST_KEYS:
+            nested = step.get(key)
+            if isinstance(nested, list):
+                result = find_step(nested, step_id)
+                if result is not None:
+                    return result
+        cases = step.get("cases")
+        if isinstance(cases, dict):
+            for case_steps in cases.values():
+                if isinstance(case_steps, list):
+                    result = find_step(case_steps, step_id)
+                    if result is not None:
+                        return result
+    return None
+
+
+def _all_base_step_ids(steps: list[dict[str, Any]]) -> set[str]:
+    """Collect all step IDs reachable in a step tree (excluding fan-out templates)."""
+    ids: set[str] = set()
+    for step in steps:
+        if not isinstance(step, dict):
+            continue
+        step_id = step.get("id")
+        if isinstance(step_id, str):
+            ids.add(step_id)
+        for key in _NESTED_LIST_KEYS:
+            nested = step.get(key)
+            if isinstance(nested, list):
+                ids.update(_all_base_step_ids(nested))
+        cases = step.get("cases")
+        if isinstance(cases, dict):
+            for case_steps in cases.values():
+                if isinstance(case_steps, list):
+                    ids.update(_all_base_step_ids(case_steps))
+    return ids
+
+
+def _descendant_ids(step: dict[str, Any]) -> set[str]:
+    """Return all step IDs nested inside *step* (not including *step* itself)."""
+    ids: set[str] = set()
+    for key in _NESTED_LIST_KEYS:
+        nested = step.get(key)
+        if isinstance(nested, list):
+            ids.update(_all_base_step_ids(nested))
+    cases = step.get("cases")
+    if isinstance(cases, dict):
+        for case_steps in cases.values():
+            if isinstance(case_steps, list):
+                ids.update(_all_base_step_ids(case_steps))
+    return ids
+
+
+def _check_anchor_conflicts(
+    anchor_operations: dict[str, str],
+    base_steps: list[dict[str, Any]],
+) -> list[str]:
+    """Return error messages for anchor pairs where one is an ancestor of the other.
+
+    Only flags conflicts where the ancestor's winning edit is ``replace`` or
+    ``remove`` — operations that destroy the subtree and make any descendant
+    anchor unresolvable.  Pure insert operations on an ancestor leave it intact,
+    so its descendants remain reachable regardless of processing order.
+
+    Callers should raise on any returned errors before mutating the step tree.
+    """
+    errors: list[str] = []
+    for anchor, operation in sorted(anchor_operations.items()):
+        if operation in ("insert_after", "insert_before"):
+            # Inserts leave the ancestor step intact; descendants are unaffected.
+            continue
+        location = find_step(base_steps, anchor)
+        if location is None:
+            continue  # missing anchors are reported by validate_edits
+        parent_list, idx = location
+        step = parent_list[idx]
+        conflicting = set(anchor_operations.keys()) & _descendant_ids(step)
+        for child_anchor in sorted(conflicting):
+            errors.append(
+                f"Anchor conflict: '{anchor}' is an ancestor of '{child_anchor}'. "
+                "Targeting both anchors in the same overlay set produces "
+                "order-dependent results; restructure edits to avoid nesting."
+            )
+    return errors
+
+
+def _init_sources_recursively(
+    steps: list[dict[str, Any]], sources: dict[str, str]
+) -> None:
+    """Initialize attribution sources for all base steps, recursively."""
+    for step in steps:
+        if not isinstance(step, dict):
+            continue
+        step_id = step.get("id")
+        if isinstance(step_id, str):
+            sources[step_id] = "base"
+        for key in _NESTED_LIST_KEYS:
+            nested = step.get(key)
+            if isinstance(nested, list):
+                _init_sources_recursively(nested, sources)
+        cases = step.get("cases")
+        if isinstance(cases, dict):
+            for case_steps in cases.values():
+                if isinstance(case_steps, list):
+                    _init_sources_recursively(case_steps, sources)
+
+
+def _record_sources_recursively(
+    step: dict[str, Any],
+    source: str,
+    sources: dict[str, str],
+) -> None:
+    """Record *source* for a step and all its nested child steps.
+
+    Traverses ``then``, ``else``, ``steps``, ``default``, and ``cases.*``
+    so that ``workflow resolve`` attributes every step inside a composite
+    insert or replacement to the correct overlay layer.
+    """
+    step_id = step.get("id")
+    if isinstance(step_id, str):
+        sources[step_id] = source
+    for key in _NESTED_LIST_KEYS:
+        nested = step.get(key)
+        if isinstance(nested, list):
+            for child in nested:
+                if isinstance(child, dict):
+                    _record_sources_recursively(child, source, sources)
+    cases = step.get("cases")
+    if isinstance(cases, dict):
+        for case_steps in cases.values():
+            if isinstance(case_steps, list):
+                for child in case_steps:
+                    if isinstance(child, dict):
+                        _record_sources_recursively(child, source, sources)
+
+
+def _remove_sources_recursively(
+    step: dict[str, Any],
+    sources: dict[str, str],
+) -> None:
+    """Remove source entries for a step and all its nested child steps.
+
+    Traverses the same nesting keys as ``_record_sources_recursively``.
+    """
+    step_id = step.get("id")
+    if isinstance(step_id, str) and sources.get(step_id) == "base":
+        sources.pop(step_id, None)
+    for key in _NESTED_LIST_KEYS:
+        nested = step.get(key)
+        if isinstance(nested, list):
+            for child in nested:
+                if isinstance(child, dict):
+                    _remove_sources_recursively(child, sources)
+    cases = step.get("cases")
+    if isinstance(cases, dict):
+        for case_steps in cases.values():
+            if isinstance(case_steps, list):
+                for child in case_steps:
+                    if isinstance(child, dict):
+                        _remove_sources_recursively(child, sources)
+
+
+
+def _build_attribution(
+    steps: list[dict[str, Any]],
+    sources: dict[str, str],
+) -> list[ComposedStep]:
+    """Build an ordered attribution list from the composed step tree."""
+    result: list[ComposedStep] = []
+    for step in steps:
+        if not isinstance(step, dict):
+            continue
+        step_id = step.get("id")
+        if isinstance(step_id, str):
+            result.append(ComposedStep(step_id, sources.get(step_id, "unknown")))
+        for key in _NESTED_LIST_KEYS:
+            nested = step.get(key)
+            if isinstance(nested, list):
+                result.extend(_build_attribution(nested, sources))
+        cases = step.get("cases")
+        if isinstance(cases, dict):
+            for case_steps in cases.values():
+                if isinstance(case_steps, list):
+                    result.extend(_build_attribution(case_steps, sources))
+    return result
+
+
+def _traverse_and_apply(
+    steps: list[dict[str, Any]],
+    edits_by_anchor: dict[str, list[tuple[OverlayLayer, OverlayEdit]]],
+    sources: dict[str, str],
+) -> list[dict[str, Any]]:
+    """Walk the original step tree and apply overlay edits as each step is encountered.
+
+    Edits are always resolved against the *original* structure — this function
+    traverses the unmodified list passed in, so a replacement step's new ID can
+    never be mistaken for a base anchor.  Nested lists (``then``, ``else``, etc.)
+    are recursed into only for steps that survive the edit (not for replaced
+    steps).
+
+    *edits* are expected to be in merge order (lowest priority first, highest
+    priority last); the winning edit for each anchor is ``edits[-1]``.
+    """
+    result: list[dict[str, Any]] = []
+
+    for step in steps:
+        if not isinstance(step, dict):
+            result.append(step)
+            continue
+
+        step_id = step.get("id")
+        edits = edits_by_anchor.get(step_id, []) if isinstance(step_id, str) else []
+        winning_edit = edits[-1][1] if edits else None
+
+        if winning_edit is not None and winning_edit.operation == "remove":
+            # Winning edit removes this step; ignore all other edits on this anchor.
+            # Do NOT call _remove_sources_recursively here: _build_attribution only
+            # traverses the result list, so stale sources entries for removed steps
+            # are never read.  Calling it would incorrectly pop the attribution of a
+            # *surviving* step that reuses the same ID (e.g. a replacement step
+            # introduced by a higher-priority overlay targeting a different anchor).
+            continue
+
+        # Insert before (in merge order).
+        for layer, edit in edits:
+            if edit.operation == "insert_before":
+                new_step = copy.deepcopy(edit.step)
+                _record_sources_recursively(new_step, layer.source, sources)
+                result.append(new_step)
+
+        if winning_edit is not None and winning_edit.operation == "replace":
+            winning_layer = edits[-1][0]
+            new_step = copy.deepcopy(winning_edit.step)
+            _remove_sources_recursively(step, sources)
+            _record_sources_recursively(new_step, winning_layer.source, sources)
+            result.append(new_step)
+        else:
+            # No replacement: keep this step and recurse into its nested lists.
+            for key in _NESTED_LIST_KEYS:
+                nested = step.get(key)
+                if isinstance(nested, list):
+                    step[key] = _traverse_and_apply(nested, edits_by_anchor, sources)
+            cases = step.get("cases")
+            if isinstance(cases, dict):
+                for case_key, case_steps in cases.items():
+                    if isinstance(case_steps, list):
+                        cases[case_key] = _traverse_and_apply(case_steps, edits_by_anchor, sources)
+            result.append(step)
+
+        # Insert after (highest priority closest to anchor — reversed merge order).
+        for layer, edit in reversed(edits):
+            if edit.operation == "insert_after":
+                new_step = copy.deepcopy(edit.step)
+                _record_sources_recursively(new_step, layer.source, sources)
+                result.append(new_step)
+
+    return result
+
+
+def merge_steps(
+    base_steps: list[dict[str, Any]],
+    overlays: list[OverlayLayer],
+) -> tuple[list[dict[str, Any]], list[ComposedStep]]:
+    """Apply overlays to base steps in merge order and return composed steps.
+
+    *overlays* is expected to be sorted by merge order (lowest priority first,
+    highest priority last).  The returned step list is a deep copy of the base;
+    base_steps is never mutated.
+
+    Higher-wins semantics are enforced for edits that target the same base
+    anchor: the highest-priority edit (last in *overlays*) decides the fate of
+    the anchor.  A lower-priority ``remove`` cannot prevent a higher-priority
+    ``replace`` or ``insert_*`` on the same anchor.
+    """
+    steps = copy.deepcopy(base_steps)
+    sources: dict[str, str] = {}
+    _init_sources_recursively(steps, sources)
+
+    # Group edits by anchor, preserving merge order.
+    edits_by_anchor: dict[str, list[tuple[OverlayLayer, OverlayEdit]]] = {}
+    for layer in overlays:
+        for edit in layer.overlay.edits:
+            edits_by_anchor.setdefault(edit.anchor, []).append((layer, edit))
+
+    # Raise early for non-remove edits that target anchors not present in the base.
+    # Overlays always apply to the original tree; they cannot target steps introduced
+    # by other overlays.
+    base_ids = _all_base_step_ids(base_steps)
+    for anchor, anchor_edits in edits_by_anchor.items():
+        winning_op = anchor_edits[-1][1].operation
+        if winning_op != "remove" and anchor not in base_ids:
+            raise ValueError(f"Anchor '{anchor}' not found in workflow steps.")
+
+    # Reject edits that target anchors with a parent/descendant relationship when
+    # the ancestor edit replaces or removes its subtree — those produce
+    # order-dependent results.  Pure insert edits on an ancestor are safe because
+    # the ancestor step (and its descendants) remain intact.
+    anchor_winning_ops = {
+        anchor: anchor_edits[-1][1].operation
+        for anchor, anchor_edits in edits_by_anchor.items()
+    }
+    anchor_conflicts = _check_anchor_conflicts(anchor_winning_ops, base_steps)
+    if anchor_conflicts:
+        raise ValueError(
+            "Overlay anchor conflict(s) detected:\n  - " + "\n  - ".join(anchor_conflicts)
+        )
+
+    # Apply all overlay edits via a single-pass traversal of the original tree.
+    # Each edit is resolved against the original step structure, so a replacement
+    # step's new ID can never be mistaken for a base anchor in a later edit group.
+    result = _traverse_and_apply(steps, edits_by_anchor, sources)
+
+    attribution = _build_attribution(result, sources)
+    return result, attribution
+
+
+def validate_edits(
+    edits: list[OverlayEdit],
+    base_step_ids: set[str],
+) -> list[str]:
+    """Validate overlay edits against a set of known base step IDs.
+
+    Returns a list of human-readable error messages.  Does not raise.
+    """
+    errors: list[str] = []
+    for idx, edit in enumerate(edits):
+        if edit.operation not in VALID_OPERATIONS:
+            errors.append(f"Edit {idx}: invalid operation {edit.operation!r}.")
+            continue
+        if edit.anchor not in base_step_ids:
+            errors.append(
+                f"Edit {idx}: anchor '{edit.anchor}' does not match any base step id."
+            )
+        if edit.operation == "remove":
+            if edit.step is not None:
+                errors.append(f"Edit {idx}: 'remove' must not include a step.")
+            continue
+        if not isinstance(edit.step, dict):
+            errors.append(f"Edit {idx}: '{edit.operation}' requires a step mapping.")
+            continue
+        step_id = edit.step.get("id")
+        if not isinstance(step_id, str) or not step_id:
+            errors.append(f"Edit {idx}: step is missing required 'id'.")
+            continue
+        if ":" in step_id:
+            errors.append(
+                f"Edit {idx}: step id {step_id!r} contains ':' which is reserved "
+                "for engine-generated nested IDs."
+            )
+    return errors

src/specify_cli/workflows/overlays/schema.py

@@ -0,0 +1,176 @@
+"""Workflow overlay schema — dataclasses and validation for overlay manifests."""
+
+from __future__ import annotations
+
+import re
+from dataclasses import dataclass
+from typing import Any, Literal
+
+from ...extensions import normalize_priority
+
+# Safe single-segment identifiers: no path separators, no traversal, no dots.
+_SAFE_ID_PATTERN = re.compile(r"^[a-z0-9](?:[a-z0-9-]*[a-z0-9])?$")
+_RESERVED_OVERLAY_WORKFLOW_IDS: frozenset[str] = frozenset({"overlays"})
+_RESERVED_WORKFLOW_IDS: frozenset[str] = frozenset({"overlays", "runs", "steps"})
+
+VALID_OPERATIONS = frozenset({"insert_after", "insert_before", "replace", "remove"})
+
+# Map shorthand keys to operation names.
+_SHORTHAND_OPERATION_KEYS: frozenset[str] = VALID_OPERATIONS
+
+
+@dataclass(frozen=True)
+class OverlayEdit:
+    """A single edit operation on a workflow step list."""
+
+    operation: Literal["insert_after", "insert_before", "replace", "remove"]
+    anchor: str
+    step: dict[str, Any] | None = None
+
+
+@dataclass
+class Overlay:
+    """A declared overlay (one YAML file)."""
+
+    id: str
+    extends: str
+    edits: list[OverlayEdit]
+    priority: int = 10
+    enabled: bool = True
+
+
+def _validate_safe_id(
+    value: str,
+    field_name: str,
+    allow_reserved: bool = False,
+    reserved_ids: frozenset[str] = _RESERVED_OVERLAY_WORKFLOW_IDS,
+) -> str | None:
+    """Return an error message if *value* is not a safe path segment ID."""
+    if not isinstance(value, str) or not value:
+        return f"Overlay '{field_name}' is required and must be a non-empty string."
+    if not _SAFE_ID_PATTERN.fullmatch(value):
+        return (
+            f"Overlay '{field_name}' {value!r} contains invalid characters; "
+            "only lowercase letters, digits, and hyphens are allowed."
+        )
+    if not allow_reserved and value in reserved_ids:
+        return f"Overlay '{field_name}' {value!r} is reserved."
+    return None
+
+
+def _parse_edit(edit_raw: dict[str, Any], idx: int) -> tuple[OverlayEdit | None, str | None]:
+    """Parse a single edit dict into an OverlayEdit or an error string."""
+    shorthand_keys = [key for key in _SHORTHAND_OPERATION_KEYS if key in edit_raw]
+    has_operation = "operation" in edit_raw
+
+    operation: str | None = None
+    anchor: Any = None
+
+    if shorthand_keys and has_operation:
+        return None, (
+            f"Edit at index {idx} mixes shorthand operation key "
+            f"({shorthand_keys[0]!r}) with explicit 'operation' field."
+        )
+
+    if len(shorthand_keys) > 1:
+        return None, (
+            f"Edit at index {idx} has multiple operation keys: "
+            f"{', '.join(repr(k) for k in shorthand_keys)}."
+        )
+
+    if shorthand_keys:
+        operation = shorthand_keys[0]
+        anchor = edit_raw[operation]
+    elif has_operation:
+        operation = edit_raw.get("operation")
+        anchor = edit_raw.get("anchor")
+    else:
+        return None, f"Edit at index {idx} has no operation; expected one of {sorted(VALID_OPERATIONS)}."
+
+    if operation not in VALID_OPERATIONS:
+        return None, f"Edit at index {idx} has invalid operation {operation!r}."
+
+    if not isinstance(anchor, str) or not anchor:
+        return None, f"Edit at index {idx} has invalid 'anchor'."
+
+    step = edit_raw.get("step")
+    if operation == "remove":
+        if step is not None:
+            return None, f"Edit at index {idx} ('remove') must not include 'step'."
+        return OverlayEdit(operation=operation, anchor=anchor), None
+
+    if not isinstance(step, dict):
+        return None, f"Edit at index {idx} ('{operation}') requires 'step' mapping."
+    step_id = step.get("id")
+    if not isinstance(step_id, str) or not step_id:
+        return None, f"Edit at index {idx} step is missing required 'id'."
+    if ":" in step_id:
+        return None, (
+            f"Edit at index {idx} step id {step_id!r} contains ':' "
+            "which is reserved for engine-generated nested IDs."
+        )
+    return OverlayEdit(operation=operation, anchor=anchor, step=step), None
+
+
+def validate_overlay_yaml(data: dict[str, Any]) -> tuple[Overlay | None, list[str]]:
+    """Validate an overlay manifest dict and return (Overlay, errors).
+
+    Errors are returned as a list of strings; validation never raises.
+    """
+    errors: list[str] = []
+
+    if not isinstance(data, dict):
+        return None, ["Overlay manifest must be a mapping."]
+
+    overlay_id = data.get("id")
+    if err := _validate_safe_id(overlay_id, "id"):
+        errors.append(err)
+        overlay_id = ""
+
+    extends = data.get("extends")
+    if err := _validate_safe_id(
+        extends,
+        "extends",
+        reserved_ids=_RESERVED_WORKFLOW_IDS,
+    ):
+        errors.append(err)
+        extends = ""
+
+    priority = normalize_priority(data.get("priority", 10))
+
+    edits_raw = data.get("edits")
+    edits: list[OverlayEdit] = []
+    if not isinstance(edits_raw, list):
+        errors.append("Overlay 'edits' is required and must be a list.")
+    elif not edits_raw:
+        errors.append("Overlay 'edits' must be a non-empty list.")
+    else:
+        for idx, edit_raw in enumerate(edits_raw):
+            if not isinstance(edit_raw, dict):
+                errors.append(f"Edit at index {idx} must be a mapping.")
+                continue
+            edit, err = _parse_edit(edit_raw, idx)
+            if err:
+                errors.append(err)
+                continue
+            if edit is not None:
+                edits.append(edit)
+
+    enabled = data.get("enabled", True)
+    if not isinstance(enabled, bool):
+        errors.append("Overlay 'enabled' must be a boolean.")
+        enabled = bool(enabled)
+
+    if errors:
+        return None, errors
+
+    return (
+        Overlay(
+            id=overlay_id,
+            extends=extends,
+            priority=priority,
+            edits=edits,
+            enabled=enabled,
+        ),
+        [],
+    )

src/specify_cli/workflows/steps/command/__init__.py

@@ -30,6 +30,21 @@ class CommandStep(StepBase):
 
     def execute(self, config: dict[str, Any], context: StepContext) -> StepResult:
         command = config.get("command", "")
+        # validate() rejects a non-string 'command', but the engine does not
+        # auto-validate before execute(); an unvalidated run would pass the value
+        # to build_command_invocation() (via _try_dispatch) and crash there with a
+        # raw AttributeError (command_name.startswith(...) on a list/int/None).
+        # Fail the step with the same contract error instead, mirroring the
+        # 'input'/'options' guards below.
+        if not isinstance(command, str):
+            return StepResult(
+                status=StepStatus.FAILED,
+                error=(
+                    f"Command step {config.get('id', '?')!r}: 'command' must be a "
+                    f"string, got {type(command).__name__}."
+                ),
+            )
+
         input_data = config.get("input", {})
         # validate() rejects a non-mapping input, but the engine does not
         # auto-validate before execute(); a workflow that skipped validation can
@@ -51,16 +66,52 @@ class CommandStep(StepBase):
         for key, value in input_data.items():
             resolved_input[key] = evaluate_expression(value, context)
 
-        # Resolve integration (step → workflow default → project default)
-        integration = config.get("integration") or context.default_integration
+        # Resolve integration (step → workflow default → project default).
+        # Fall back to the workflow default ONLY for a genuinely-unset value
+        # (missing / YAML-null / empty string). A ``config.get(...) or ...``
+        # would also swallow a falsey *non-string* ([], {}, 0, False), coercing
+        # it to the default before the guard below runs — so on an unvalidated
+        # execute() such a step would silently dispatch with the configured
+        # default instead of failing. Fall through instead, so every non-string
+        # reaches the type guard.
+        integration = config.get("integration")
+        if integration is None or integration == "":
+            integration = context.default_integration
         if integration and isinstance(integration, str) and "{{" in integration:
             integration = evaluate_expression(integration, context)
 
-        # Resolve model
-        model = config.get("model") or context.default_model
+        # Resolve model (same fallback rationale as 'integration' above).
+        model = config.get("model")
+        if model is None or model == "":
+            model = context.default_model
         if model and isinstance(model, str) and "{{" in model:
             model = evaluate_expression(model, context)
 
+        # A non-string integration/model — a literal list/dict/number that
+        # skipped validation, an unvalidated workflow-level default, or an
+        # expression that resolved to one — crashes downstream: get_integration()
+        # uses the value as a dict key (raw TypeError on an unhashable list/dict,
+        # even on a *validated* run) and build_exec_args() feeds model into the
+        # CLI argv. Fail the step with the contract error rather than taking down
+        # the whole run, mirroring the 'input'/'options' guards above. ``None``
+        # stays valid — it means "unset" and falls back to dispatch-not-possible.
+        if integration is not None and not isinstance(integration, str):
+            return StepResult(
+                status=StepStatus.FAILED,
+                error=(
+                    f"Command step {config.get('id', '?')!r}: 'integration' must "
+                    f"be a string, got {type(integration).__name__}."
+                ),
+            )
+        if model is not None and not isinstance(model, str):
+            return StepResult(
+                status=StepStatus.FAILED,
+                error=(
+                    f"Command step {config.get('id', '?')!r}: 'model' must be a "
+                    f"string, got {type(model).__name__}."
+                ),
+            )
+
         # Merge options (workflow defaults ← step overrides)
         options = dict(context.default_options)
         step_options = config.get("options", {})
@@ -138,7 +189,11 @@ class CommandStep(StepBase):
         not possible (integration not found, CLI not installed, or
         dispatch not supported).
         """
-        if not integration_key:
+        if not integration_key or not isinstance(integration_key, str):
+            # A non-string integration (a list/dict/expression that resolved to
+            # one) would raise TypeError: unhashable type from get_integration's
+            # dict lookup below and abort the whole run. Treat it as "not
+            # dispatchable" so execute() falls through to its FAILED StepResult.
             return None
 
         try:
@@ -179,6 +234,17 @@ class CommandStep(StepBase):
             errors.append(
                 f"Command step {config.get('id', '?')!r} is missing 'command' field."
             )
+        elif not isinstance(config["command"], str):
+            # execute() passes 'command' straight to the integration's
+            # build_command_invocation(), which does command_name.startswith(...);
+            # a non-string (null, list, int) crashes there with a raw
+            # AttributeError once dispatch is attempted. Reject it at validation,
+            # mirroring the prompt-step 'prompt' and shell-step 'run' type checks.
+            # An expression like "{{ ... }}" is still a str, so it stays valid.
+            errors.append(
+                f"Command step {config.get('id', '?')!r}: 'command' must be a "
+                f"string, got {type(config['command']).__name__}."
+            )
         # execute() iterates input.items() and options.update(step_options); a
         # non-mapping here would raise at run time. Validate the shape like the
         # sibling steps (switch 'cases', fan-out 'step') so it is reported, not
@@ -191,4 +257,23 @@ class CommandStep(StepBase):
             errors.append(
                 f"Command step {config.get('id', '?')!r}: 'options' must be a mapping."
             )
+        # execute() passes 'integration' to get_integration(), which uses it as a
+        # dict key — a non-string (list/dict) raises a raw TypeError (unhashable),
+        # even on a validated run — and feeds 'model' into the CLI argv. Reject a
+        # literal non-string here, mirroring the sibling type checks. ``None``
+        # (an explicit ``integration:``/``model:`` YAML null) means "inherit the
+        # workflow default" and stays valid; an expression like "{{ ... }}" is
+        # still a str, so it stays valid too.
+        integration = config.get("integration")
+        if integration is not None and not isinstance(integration, str):
+            errors.append(
+                f"Command step {config.get('id', '?')!r}: 'integration' must be a "
+                f"string, got {type(integration).__name__}."
+            )
+        model = config.get("model")
+        if model is not None and not isinstance(model, str):
+            errors.append(
+                f"Command step {config.get('id', '?')!r}: 'model' must be a "
+                f"string, got {type(model).__name__}."
+            )
         return errors

src/specify_cli/workflows/steps/fan_in/__init__.py

@@ -42,6 +42,28 @@ class FanInStep(StepBase):
                 output={"results": []},
             )
 
+        # A non-string entry can never match a real step id. An unhashable one
+        # (a list/dict from a YAML indentation slip like ``wait_for: [[a, b]]``)
+        # crashes the whole run at ``context.steps.get(step_id, ...)`` below with
+        # a raw TypeError; a hashable-but-non-string one (``wait_for: [123]``)
+        # silently joins an empty ``{}`` and still reports COMPLETED — the exact
+        # "silent empty result + COMPLETED" wiring bug the whole-list guard above
+        # and the engine's fan-in validation (engine.py) both reject. The engine
+        # does not auto-validate step config, so fail this step loudly on an
+        # unvalidated run too, using the engine's phrasing.
+        bad_entries = [w for w in wait_for if not isinstance(w, str)]
+        if bad_entries:
+            first = bad_entries[0]
+            return StepResult(
+                status=StepStatus.FAILED,
+                error=(
+                    f"Fan-in step {config.get('id', '?')!r}: 'wait_for' entries "
+                    f"must be step-id strings, got {type(first).__name__} "
+                    f"({first!r})."
+                ),
+                output={"results": []},
+            )
+
         # Collect results from referenced steps
         results = []
         for step_id in wait_for:

src/specify_cli/workflows/steps/fan_out/__init__.py

@@ -12,9 +12,10 @@ class FanOutStep(StepBase):
     """Dispatch a step template for each item in a collection.
 
     The engine executes the nested ``step:`` template once per item,
-    setting ``context.item`` for each iteration.  Execution is
-    currently sequential; ``max_concurrency`` is accepted but not
-    enforced.
+    setting ``context.item`` for each iteration.  ``max_concurrency``
+    controls parallelism: ``<= 1`` (the default) runs items
+    sequentially, while ``> 1`` runs up to that many items concurrently
+    on a bounded thread pool (see ``WorkflowEngine._run_fan_out``).
     """
 
     type_key = "fan-out"
@@ -25,6 +26,33 @@ class FanOutStep(StepBase):
         max_concurrency = config.get("max_concurrency", 1)
         step_template = config.get("step", {})
 
+        # The engine does not auto-validate step config (see
+        # ``WorkflowEngine.load_workflow``). On a COMPLETED fan-out it reads the
+        # ``step_template`` back out and, when it is truthy, calls
+        # ``template.get("id", ...)`` in ``_run_fan_out``. A truthy non-mapping
+        # ``step`` (a scalar or list authoring mistake) would crash the whole
+        # run with AttributeError there — the engine invokes ``execute`` and
+        # ``_run_fan_out`` with no surrounding try/except. ``validate`` already
+        # rejects a non-mapping ``step``; fail this step loudly on an
+        # unvalidated run instead, mirroring the ``items`` guard below. An empty
+        # or absent ``step`` defaults to ``{}`` (falsy) and the engine's
+        # ``if template and items`` skips fan-out, so it stays valid here.
+        if not isinstance(step_template, dict):
+            return StepResult(
+                status=StepStatus.FAILED,
+                error=(
+                    f"Fan-out step {config.get('id', '?')!r}: 'step' must be a "
+                    f"mapping (nested step template), got "
+                    f"{type(step_template).__name__}."
+                ),
+                output={
+                    "items": [],
+                    "max_concurrency": max_concurrency,
+                    "step_template": {},
+                    "item_count": 0,
+                },
+            )
+
         if not isinstance(items, list):
             # A non-list here is a wiring error (the expression did not
             # resolve to a collection); silently fanning out over zero
@@ -66,8 +94,13 @@ class FanOutStep(StepBase):
                 f"Fan-out step {config.get('id', '?')!r} is missing "
                 f"'step' field (nested step template)."
             )
-        step = config.get("step")
-        if step is not None and not isinstance(step, dict):
+        elif not isinstance(config["step"], dict):
+            # A present-but-non-mapping ``step`` (including an explicit
+            # ``step: null``) is an authoring mistake. ``config.get("step", {})``
+            # in ``execute`` only substitutes the ``{}`` default for an *absent*
+            # key, so an explicit ``None`` reaches the runtime guard and FAILS
+            # the step. Reject it here too so a workflow cannot pass validation
+            # and then fail during execution.
             errors.append(
                 f"Fan-out step {config.get('id', '?')!r}: 'step' must be a mapping."
             )

src/specify_cli/workflows/steps/gate/__init__.py

@@ -43,6 +43,35 @@ class GateStep(StepBase):
         options = config.get("options", ["approve", "reject"])
         on_reject = config.get("on_reject", "abort")
 
+        # ``validate`` rejects a non-list (or empty) ``options``, and requires
+        # every option to be a string, but the engine does not auto-validate
+        # before ``execute``. An unvalidated run with a scalar/dict/None
+        # ``options`` would otherwise reach ``_prompt`` and crash the whole run
+        # with a raw ``TypeError`` (``enumerate``/``len`` on a non-iterable) or
+        # ``KeyError`` (indexing a dict); a non-string option would crash at the
+        # ``choice.lower()`` reject check with ``AttributeError``. Fail this step
+        # loudly instead — mirroring the switch 'cases' and command 'input'
+        # guards. Checked before the non-TTY short-circuit so the error surfaces
+        # in CI too, rather than PAUSING and crashing later on interactive resume.
+        if (
+            not isinstance(options, list)
+            or not options
+            or not all(isinstance(o, str) for o in options)
+        ):
+            return StepResult(
+                status=StepStatus.FAILED,
+                error=(
+                    f"Gate step {config.get('id', '?')!r}: 'options' must be a "
+                    f"non-empty list of strings, got {type(options).__name__}."
+                ),
+                output={
+                    "message": message,
+                    "options": options,
+                    "on_reject": on_reject,
+                    "choice": None,
+                },
+            )
+
         show_file = config.get("show_file")
         if isinstance(show_file, str) and "{{" in show_file:
             show_file = evaluate_expression(show_file, context)
@@ -139,7 +168,11 @@ class GateStep(StepBase):
             except (EOFError, KeyboardInterrupt):
                 print()
                 return options[-1]  # default to last (usually reject)
-            if raw.isdigit() and 1 <= int(raw) <= len(options):
+            # isdecimal() (not isdigit()): int() accepts exactly the decimal-digit
+            # set, whereas isdigit() also returns True for superscripts/subscripts
+            # (e.g. "²") that int() then rejects with ValueError — crashing
+            # this interactive loop.
+            if raw.isdecimal() and 1 <= int(raw) <= len(options):
                 return options[int(raw) - 1]
             # Also accept the option name directly
             if raw.lower() in [o.lower() for o in options]:

src/specify_cli/workflows/steps/prompt/__init__.py

@@ -42,16 +42,52 @@ class PromptStep(StepBase):
         if not isinstance(prompt, str):
             prompt = str(prompt)
 
-        # Resolve integration (step → workflow default)
-        integration = config.get("integration") or context.default_integration
+        # Resolve integration (step → workflow default).
+        # Fall back to the workflow default ONLY for a genuinely-unset value
+        # (missing / YAML-null / empty string). A ``config.get(...) or ...``
+        # would also swallow a falsey *non-string* ([], {}, 0, False), coercing
+        # it to the default before the guard below runs — so on an unvalidated
+        # execute() such a step would silently dispatch with the configured
+        # default instead of failing. Fall through instead, so every non-string
+        # reaches the type guard.
+        integration = config.get("integration")
+        if integration is None or integration == "":
+            integration = context.default_integration
         if integration and isinstance(integration, str) and "{{" in integration:
             integration = evaluate_expression(integration, context)
 
-        # Resolve model
-        model = config.get("model") or context.default_model
+        # Resolve model (same fallback rationale as 'integration' above).
+        model = config.get("model")
+        if model is None or model == "":
+            model = context.default_model
         if model and isinstance(model, str) and "{{" in model:
             model = evaluate_expression(model, context)
 
+        # A non-string integration/model — a literal list/dict/number that
+        # skipped validation, an unvalidated workflow-level default, or an
+        # expression that resolved to one — crashes downstream: get_integration()
+        # uses the value as a dict key (raw TypeError on an unhashable list/dict,
+        # even on a *validated* run) and build_exec_args() feeds model into the
+        # CLI argv. Fail the step with the contract error rather than taking down
+        # the whole run. ``None`` stays valid — it means "unset" and falls back
+        # to dispatch-not-possible.
+        if integration is not None and not isinstance(integration, str):
+            return StepResult(
+                status=StepStatus.FAILED,
+                error=(
+                    f"Prompt step {config.get('id', '?')!r}: 'integration' must "
+                    f"be a string, got {type(integration).__name__}."
+                ),
+            )
+        if model is not None and not isinstance(model, str):
+            return StepResult(
+                status=StepStatus.FAILED,
+                error=(
+                    f"Prompt step {config.get('id', '?')!r}: 'model' must be a "
+                    f"string, got {type(model).__name__}."
+                ),
+            )
+
         # Attempt CLI dispatch
         dispatch_result = self._try_dispatch(
             prompt, integration, model, context
@@ -102,7 +138,10 @@ class PromptStep(StepBase):
         context: StepContext,
     ) -> dict[str, Any] | None:
         """Dispatch *prompt* directly through the integration CLI."""
-        if not integration_key or not prompt:
+        if not integration_key or not isinstance(integration_key, str) or not prompt:
+            # A non-string integration would raise TypeError: unhashable type
+            # from get_integration's dict lookup and abort the run; treat it as
+            # not dispatchable so execute() falls through to its FAILED result.
             return None
 
         try:
@@ -160,4 +199,35 @@ class PromptStep(StepBase):
             errors.append(
                 f"Prompt step {config.get('id', '?')!r} is missing 'prompt' field."
             )
+        elif not isinstance(config["prompt"], str):
+            # execute() str()-coerces prompt and dispatches it to the
+            # integration CLI, so a null or list 'prompt' would send the Python
+            # repr ('None', "['review', 'this']") to the model as instructions —
+            # silently wrong, with no error. Reject non-strings at validation,
+            # mirroring the shell-step 'run' and command-step input/options type
+            # checks. An expression like "{{ ... }}" is still a str, so it stays
+            # valid.
+            errors.append(
+                f"Prompt step {config.get('id', '?')!r}: 'prompt' must be a "
+                f"string, got {type(config['prompt']).__name__}."
+            )
+        # execute() passes 'integration' to get_integration(), which uses it as a
+        # dict key — a non-string (list/dict) raises a raw TypeError (unhashable),
+        # even on a validated run — and feeds 'model' into the CLI argv. Reject a
+        # literal non-string here, mirroring the 'prompt' check above. ``None``
+        # (an explicit ``integration:``/``model:`` YAML null) means "inherit the
+        # workflow default" and stays valid; an expression like "{{ ... }}" is
+        # still a str, so it stays valid too.
+        integration = config.get("integration")
+        if integration is not None and not isinstance(integration, str):
+            errors.append(
+                f"Prompt step {config.get('id', '?')!r}: 'integration' must be a "
+                f"string, got {type(integration).__name__}."
+            )
+        model = config.get("model")
+        if model is not None and not isinstance(model, str):
+            errors.append(
+                f"Prompt step {config.get('id', '?')!r}: 'model' must be a "
+                f"string, got {type(model).__name__}."
+            )
         return errors

templates/commands/plan.md

@@ -11,6 +11,7 @@ handoffs:
 scripts:
   sh: scripts/bash/setup-plan.sh --json
   ps: scripts/powershell/setup-plan.ps1 -Json
+  py: scripts/python/setup_plan.py --json
 ---
 
 ## User Input

templates/commands/tasks.md

@@ -12,6 +12,7 @@ handoffs:
 scripts:
   sh: scripts/bash/setup-tasks.sh --json
   ps: scripts/powershell/setup-tasks.ps1 -Json
+  py: scripts/python/setup_tasks.py --json
 ---
 
 ## User Input

tests/contract/test_catalog_schema.py

@@ -5,6 +5,8 @@ built-in, install policy gating, payload parsing.
 """
 from __future__ import annotations
 
+import json
+import tomllib
 from pathlib import Path
 
 import yaml
@@ -37,9 +39,46 @@ def test_builtin_default_stack_when_no_config(tmp_path: Path):
     assert ids == ["default", "community"]
     assert sources[0].install_policy is InstallPolicy.INSTALL_ALLOWED
     assert sources[1].install_policy is InstallPolicy.DISCOVERY_ONLY
+    assert sources[1].priority == 20
     assert all(s.scope is Scope.BUILTIN for s in sources)
 
 
+def test_non_list_catalogs_raises_actionable_error(tmp_path: Path):
+    """A scalar ``catalogs:`` value raises a clean BundlerError, not a raw
+    'int object is not iterable' TypeError — matching what the sibling reader
+    (bundle catalog list) already reports for the same file."""
+    make_project(tmp_path)
+    (tmp_path / ".specify" / "bundle-catalogs.yml").write_text(
+        "catalogs: 5\n", encoding="utf-8"
+    )
+    with pytest.raises(BundlerError, match="must be a list"):
+        load_source_stack(tmp_path)
+
+
+@pytest.mark.parametrize("value", ["false", "0", "''", "{}"])
+def test_falsy_non_list_catalogs_still_raises(tmp_path: Path, value: str):
+    """A *falsy* non-list ``catalogs:`` value (false/0/''/{}) must also raise —
+    only an absent/``None`` value means "nothing to merge". A plain falsy check
+    would silently swallow these, diverging from the sibling reader."""
+    make_project(tmp_path)
+    (tmp_path / ".specify" / "bundle-catalogs.yml").write_text(
+        f"catalogs: {value}\n", encoding="utf-8"
+    )
+    with pytest.raises(BundlerError, match="must be a list"):
+        load_source_stack(tmp_path)
+
+
+@pytest.mark.parametrize("body", ["catalogs:\n", "catalogs: []\n"])
+def test_absent_or_empty_catalogs_is_noop(tmp_path: Path, body: str):
+    """An absent (``None``) or empty-list ``catalogs:`` is valid: it contributes
+    no project sources and falls back to the built-in default stack."""
+    make_project(tmp_path)
+    (tmp_path / ".specify" / "bundle-catalogs.yml").write_text(body, encoding="utf-8")
+    # Does not raise; still yields the built-in defaults.
+    sources = load_source_stack(tmp_path)
+    assert len(sources) > 0
+
+
 def test_project_config_overrides_same_id(tmp_path: Path):
     make_project(tmp_path)
     config = {
@@ -95,6 +134,29 @@ def test_builtin_default_stack_constant_shape():
     assert ids == {"default", "community"}
 
 
+def test_repository_community_bundle_catalog_matches_contract():
+    catalog_path = Path(__file__).parents[2] / "bundles" / "catalog.community.json"
+    payload = json.loads(catalog_path.read_text(encoding="utf-8"))
+
+    assert payload["schema_version"] == "1.0"
+    assert payload["catalog_url"].endswith("/bundles/catalog.community.json")
+    entries = load_catalog_payload(payload)
+    assert all(entry.verified is False for entry in entries.values())
+
+
+def test_wheel_packages_community_bundle_catalog():
+    repo_root = Path(__file__).parents[2]
+    with (repo_root / "pyproject.toml").open("rb") as pyproject_file:
+        pyproject = tomllib.load(pyproject_file)
+
+    force_include = pyproject["tool"]["hatch"]["build"]["targets"]["wheel"][
+        "force-include"
+    ]
+    assert force_include["bundles/catalog.community.json"] == (
+        "specify_cli/core_pack/bundles/catalog.community.json"
+    )
+
+
 def test_catalog_entry_rejects_string_tags():
     from specify_cli.bundler.models.catalog import CatalogEntry
 

tests/contract/test_manifest_schema.py

@@ -124,3 +124,13 @@ def test_string_mcp_rejected_not_split_per_character():
     data["requires"]["mcp"] = "github"
     with pytest.raises(BundlerError, match="'requires.mcp' must be a list of strings"):
         BundleManifest.from_dict(data)
+
+
+def test_string_integration_rejected_not_silently_dropped():
+    # A present-but-non-mapping 'integration' (a bare string) was silently
+    # dropped, leaving the bundle wrongly integration-agnostic. Reject it like
+    # the sibling requires/provides mapping fields.
+    data = valid_manifest_dict()
+    data["integration"] = "copilot"
+    with pytest.raises(BundlerError, match="'integration' must be a mapping when present"):
+        BundleManifest.from_dict(data)

tests/extensions/git/test_git_extension.py

@@ -698,6 +698,22 @@ class TestCreateFeaturePowerShell:
         assert rt.returncode == 0, rt.stderr
         assert "HAS_GIT" not in rt.stdout
 
+    def test_persist_hint_matches_twins(self, tmp_path: Path):
+        """The non-JSON SPECIFY_FEATURE hint must use the '# To persist in your
+        shell: $env:SPECIFY_FEATURE = '<name>' form — matching the core
+        create-new-feature.ps1 twin and the bash/python twins of this script —
+        not the old 'environment variable set to:' wording (the env var is only
+        set in this child process, so the actionable output is the persist hint)."""
+        project = _setup_project(tmp_path)
+        result = _run_pwsh(
+            "create-new-feature-branch.ps1", project,
+            "-ShortName", "persist", "Persist hint feature",
+        )
+        assert result.returncode == 0, result.stderr
+        assert "# To persist in your shell:" in result.stdout
+        assert "$env:SPECIFY_FEATURE = '001-persist'" in result.stdout
+        assert "environment variable set to:" not in result.stdout
+
     def test_help_documents_branch_prefix(self, tmp_path: Path):
         """-Help documents both template config knobs."""
         project = _setup_project(tmp_path)

tests/extensions/test_update_agent_context_feature_json.py

@@ -11,7 +11,6 @@ import pytest
 
 from tests.conftest import requires_bash
 from tests.extensions.test_extension_agent_context import (
-    BASH,
     POWERSHELL,
     _bash_posix_path,
     _run_bash_agent_context_script,

tests/integration/test_bundler_catalog_stack.py

@@ -38,6 +38,27 @@ def test_resolve_prefers_highest_precedence_source():
     assert resolved.install_allowed is False
 
 
+def test_explicit_catalog_shadows_builtin_community_at_default_priority():
+    sources = [
+        _source("community", 20, "discovery-only"),
+        _source("explicit", 10, "install-allowed"),
+    ]
+    payloads = {
+        "community": catalog_payload({
+            "shared": catalog_entry_dict("shared", version="1.0.0"),
+        }),
+        "explicit": catalog_payload({
+            "shared": catalog_entry_dict("shared", version="2.0.0"),
+        }),
+    }
+
+    resolved = _stack(sources, payloads).resolve("shared")
+
+    assert resolved.source.id == "explicit"
+    assert resolved.entry.version == "2.0.0"
+    assert resolved.install_allowed is True
+
+
 def test_resolve_unknown_bundle_errors():
     stack = _stack(
         [_source("only", 1, "install-allowed")],

tests/integration/test_bundler_offline.py

@@ -31,6 +31,25 @@ def test_builtin_catalog_resolves_offline():
     assert stack.search() == []
 
 
+def test_builtin_community_catalog_resolves_from_packaged_snapshot_offline():
+    fetcher = make_catalog_fetcher(allow_network=False)
+    source = _src(
+        "community",
+        "builtin://community",
+        priority=20,
+        policy="discovery-only",
+    )
+    payload = fetcher(source)
+    stack = CatalogStack([source], fetcher)
+
+    assert isinstance(payload.get("bundles"), dict)
+    assert all(
+        result.source.id == "community" and not result.install_allowed
+        for result in stack.search()
+    )
+    assert stack.sources[0].install_allowed is False
+
+
 def test_file_catalog_resolves_offline(tmp_path: Path):
     catalog_path = tmp_path / "catalog.json"
     write_catalog_file(catalog_path, {"demo": catalog_entry_dict("demo")})

tests/integrations/test_base.py

@@ -1,6 +1,8 @@
 """Tests for IntegrationOption, IntegrationBase, MarkdownIntegration, and primitives."""
 
+import shlex
 import sys
+from types import SimpleNamespace
 
 import pytest
 
@@ -216,6 +218,42 @@ class TestBuildCommandInvocation:
         i = get_integration("codex")
         assert i.build_command_invocation("speckit.git.commit", "fix typo") == "/speckit-git-commit fix typo"
 
+    def test_forge_core_command_hyphenated(self):
+        """Forge installs hyphenated slash-commands (/speckit-<name>), so the
+        dispatch invocation must be hyphenated too — not the dotted default it
+        would inherit from MarkdownIntegration."""
+        from specify_cli.integrations import get_integration
+        i = get_integration("forge")
+        assert i.build_command_invocation("speckit.plan") == "/speckit-plan"
+        assert i.build_command_invocation("plan") == "/speckit-plan"
+
+    def test_forge_extension_command_hyphenated(self):
+        from specify_cli.integrations import get_integration
+        i = get_integration("forge")
+        assert i.build_command_invocation("speckit.git.commit") == "/speckit-git-commit"
+        assert (
+            i.build_command_invocation("speckit.git.commit", "fix typo")
+            == "/speckit-git-commit fix typo"
+        )
+
+    def test_cline_core_command_hyphenated(self):
+        """Cline installs hyphenated slash-commands (/speckit-<name>), so the
+        dispatch invocation must be hyphenated too — not the dotted default it
+        would inherit from MarkdownIntegration."""
+        from specify_cli.integrations import get_integration
+        i = get_integration("cline")
+        assert i.build_command_invocation("speckit.plan") == "/speckit-plan"
+        assert i.build_command_invocation("plan") == "/speckit-plan"
+
+    def test_cline_extension_command_hyphenated(self):
+        from specify_cli.integrations import get_integration
+        i = get_integration("cline")
+        assert i.build_command_invocation("speckit.git.commit") == "/speckit-git-commit"
+        assert (
+            i.build_command_invocation("speckit.git.commit", "fix typo")
+            == "/speckit-git-commit fix typo"
+        )
+
 
 class TestResolveCommandRefs:
     """Tests for __SPECKIT_COMMAND_<NAME>__ placeholder resolution."""
@@ -477,19 +515,41 @@ class TestProcessTemplatePyScriptType:
         assert ".specify/scripts/bash/check-prerequisites.sh --json" in result
         assert "python" not in result
 
+    def test_body_scripts_example_does_not_override_frontmatter(self):
+        content = (
+            "---\n"
+            "scripts:\n"
+            "  sh: scripts/bash/real.sh --json\n"
+            "---\n"
+            "Run {SCRIPT} now.\n"
+            "```yaml\n"
+            "scripts:\n"
+            "  sh: examples/not-the-command.sh\n"
+            "```\n"
+        )
+
+        result = IntegrationBase.process_template(content, "agent", "sh")
+
+        assert ".specify/scripts/bash/real.sh --json" in result
+        assert "examples/not-the-command.sh" in result
+
     def test_py_quotes_interpreter_with_spaces(self, monkeypatch):
         # An interpreter path containing whitespace (e.g. Windows
         # ``Program Files``) must be quoted so it isn't split into args.
+        interpreter = r"C:\Program Files\Python\python.exe"
         monkeypatch.setattr(
             "specify_cli.integrations.base.shutil.which", lambda name: None
         )
         monkeypatch.setattr(
             "specify_cli.integrations.base.sys.executable",
-            r"C:\Program Files\Python\python.exe",
+            interpreter,
+        )
+        monkeypatch.setattr(
+            "specify_cli.integrations.base.os", SimpleNamespace(name="posix")
         )
         result = IntegrationBase.process_template(self.CONTENT, "agent", "py")
         assert (
-            '"C:\\Program Files\\Python\\python.exe" '
+            f"{shlex.quote(interpreter)} "
             ".specify/scripts/python/check-prerequisites.py --json"
         ) in result
 
@@ -511,6 +571,39 @@ class TestProcessTemplatePyScriptType:
         )
         assert ".venv/bin/python .specify/scripts/python/check-prerequisites.py" in result
 
+    def test_setup_py_falls_back_to_platform_shell(
+        self, monkeypatch, tmp_path
+    ):
+        template = tmp_path / "fallback.md"
+        template.write_text(
+            "---\n"
+            "scripts:\n"
+            "  sh: scripts/bash/check-prerequisites.sh --json\n"
+            "  ps: scripts/powershell/check-prerequisites.ps1 -Json\n"
+            "---\n"
+            "Run {SCRIPT} now.\n",
+            encoding="utf-8",
+        )
+        integration = StubIntegration()
+        monkeypatch.setattr(
+            integration, "list_command_templates", lambda: [template]
+        )
+
+        created = integration.setup(
+            tmp_path,
+            IntegrationManifest("stub", tmp_path),
+            script_type="py",
+        )
+
+        rendered = created[0].read_text(encoding="utf-8")
+        expected = (
+            ".specify/scripts/powershell/check-prerequisites.ps1"
+            if sys.platform == "win32"
+            else ".specify/scripts/bash/check-prerequisites.sh"
+        )
+        assert "{SCRIPT}" not in rendered
+        assert expected in rendered
+
 
 class TestInstallScriptsPython:
     def _make_integration_with_scripts(self, monkeypatch, tmp_path):

tests/integrations/test_integration_bob.py

@@ -1,10 +1,927 @@
 """Tests for BobIntegration."""
 
-from .test_integration_base_markdown import MarkdownIntegrationTests
+import os
+import warnings
+
+import pytest
+import yaml
+
+from specify_cli.integrations import INTEGRATION_REGISTRY, get_integration
+from specify_cli.integrations.base import SkillsIntegration
+from specify_cli.integrations.manifest import IntegrationManifest
 
 
-class TestBobIntegration(MarkdownIntegrationTests):
-    KEY = "bob"
-    FOLDER = ".bob/"
-    COMMANDS_SUBDIR = "commands"
-    REGISTRAR_DIR = ".bob/commands"
+class TestBobIntegrationRegistration:
+    def test_registered(self):
+        assert "bob" in INTEGRATION_REGISTRY
+        assert get_integration("bob") is not None
+
+    def test_is_integration_base_not_skills_integration(self):
+        """BobIntegration extends IntegrationBase directly — not SkillsIntegration.
+
+        Bob is dual-mode (skills by default, legacy commands via
+        ``--legacy-commands``), so its skills-ness is a per-project config
+        decision resolved by the ``is_skills_mode`` hook — not a class-hierarchy
+        property.  It therefore must NOT be a ``SkillsIntegration`` (which is
+        reserved for statically skills-only agents); shared code consults
+        ``is_skills_mode(parsed_options)`` instead of ``isinstance``.
+        ``invoke_separator='-'`` is set explicitly on the class to match the
+        default (skills) layout.
+        """
+        from specify_cli.integrations.base import IntegrationBase
+        bob = get_integration("bob")
+        assert isinstance(bob, IntegrationBase)
+        assert not isinstance(bob, SkillsIntegration)
+        assert bob.invoke_separator == "-"
+
+    def test_key_and_config(self):
+        bob = get_integration("bob")
+        assert bob.key == "bob"
+        assert bob.config["folder"] == ".bob/"
+        # registrar_config mirrors the legacy commands layout so that
+        # CommandRegistrar.AGENT_CONFIGS["bob"] follows the Copilot pattern:
+        # extension registration writes to .bob/commands/ for legacy-mode
+        # projects and is skipped for skills-mode projects (skills_mode_active).
+        assert bob.config["commands_subdir"] == "commands"
+        assert bob.registrar_config["dir"] == ".bob/commands"
+        assert bob.registrar_config["extension"] == ".md"
+
+    def test_invoke_separator_is_hyphen(self):
+        """Class-level invoke_separator must be '-' so CommandRegistrar.AGENT_CONFIGS
+        generates correct /speckit-<name> refs without calling effective_invoke_separator."""
+        bob = get_integration("bob")
+        assert bob.invoke_separator == "-"
+
+
+class TestBobOptionsFlag:
+    def test_options_include_legacy_commands_flag(self):
+        bob = get_integration("bob")
+        opts = bob.options()
+        legacy_opts = [o for o in opts if o.name == "--legacy-commands"]
+        assert len(legacy_opts) == 1
+        opt = legacy_opts[0]
+        assert opt.is_flag is True
+        # Legacy must be OPT-IN (default=False) — skills are the default
+        assert opt.default is False
+
+    def test_options_include_skills_migration_flag(self):
+        """Review #3415, 4724160183, comment 1: a ``--skills`` opt-in exists as
+        the supported migration path from legacy commands to the skills layout.
+        It is distinct from the pre-skills-default ``--skills`` flag: here it
+        *forces* skills mode over on-disk auto-detection.
+        """
+        bob = get_integration("bob")
+        opts = bob.options()
+        skills_opts = [o for o in opts if o.name == "--skills"]
+        assert len(skills_opts) == 1
+        opt = skills_opts[0]
+        assert opt.is_flag is True
+        # Opt-in: disk auto-detection remains the default behavior.
+        assert opt.default is False
+
+
+class TestBobIsSkillsModeHook:
+    """The is_skills_mode hook is the single source of truth for the mode."""
+
+    def test_default_is_skills(self):
+        bob = get_integration("bob")
+        assert bob.is_skills_mode(None) is True
+        assert bob.is_skills_mode({}) is True
+
+    def test_legacy_commands_disables_skills(self):
+        bob = get_integration("bob")
+        assert bob.is_skills_mode({"legacy_commands": True}) is False
+
+    def test_existing_commands_layout_preserved_on_use(self, tmp_path):
+        """Regression (review #3415): an existing Bob 1.x project (managed
+        ``.bob/commands/speckit.*.md`` on disk, no stored ``legacy_commands``)
+        must NOT be treated as skills mode when re-resolved with a
+        project_root, so ``use``/``switch``/``upgrade`` never silently migrate
+        it to skills.
+        """
+        bob = get_integration("bob")
+        cmds = tmp_path / ".bob" / "commands"
+        cmds.mkdir(parents=True)
+        (cmds / "speckit.plan.md").write_text("# plan", encoding="utf-8")
+        # No parsed options at all — the pre-existing-install scenario.
+        assert bob.is_skills_mode(None, project_root=tmp_path) is False
+        assert bob.is_skills_mode({}, project_root=tmp_path) is False
+
+    def test_existing_skills_layout_stays_skills_on_use(self, tmp_path):
+        """A project with managed ``speckit-*`` skills resolves to skills mode."""
+        bob = get_integration("bob")
+        (tmp_path / ".bob" / "skills" / "speckit-plan").mkdir(parents=True)
+        assert bob.is_skills_mode(None, project_root=tmp_path) is True
+
+    def test_managed_commands_with_unrelated_skills_dir_stays_legacy(
+        self, tmp_path
+    ):
+        """Regression (review #3415, 4723246468): a legacy Spec Kit install
+        (managed ``.bob/commands/speckit.*.md``) that *also* carries unrelated
+        Bob 2 skills (a ``.bob/skills/`` dir with no managed ``speckit-*``
+        skills) must stay in command mode — the mere presence of a skills
+        directory is not evidence that Spec Kit is skills-based.
+        """
+        bob = get_integration("bob")
+        cmds = tmp_path / ".bob" / "commands"
+        cmds.mkdir(parents=True)
+        (cmds / "speckit.plan.md").write_text("# plan", encoding="utf-8")
+        # An unrelated (non-Spec-Kit) skill the user authored.
+        (tmp_path / ".bob" / "skills" / "my-own-skill").mkdir(parents=True)
+        assert bob.is_skills_mode(None, project_root=tmp_path) is False
+        assert bob.effective_invoke_separator(None, project_root=tmp_path) == "."
+
+    def test_managed_skills_win_when_both_layouts_present(self, tmp_path):
+        """When managed Spec Kit skills exist, skills mode wins even if a stale
+        managed command file is still on disk (upgrade leftover)."""
+        bob = get_integration("bob")
+        cmds = tmp_path / ".bob" / "commands"
+        cmds.mkdir(parents=True)
+        (cmds / "speckit.plan.md").write_text("# plan", encoding="utf-8")
+        (tmp_path / ".bob" / "skills" / "speckit-plan").mkdir(parents=True)
+        assert bob.is_skills_mode(None, project_root=tmp_path) is True
+
+    def test_fresh_project_defaults_to_skills_with_project_root(self, tmp_path):
+        """A project with no managed ``.bob/`` artifacts yet defaults to skills."""
+        bob = get_integration("bob")
+        assert bob.is_skills_mode(None, project_root=tmp_path) is True
+
+    def test_explicit_legacy_flag_wins_over_disk_layout(self, tmp_path):
+        """An explicit ``--legacy-commands`` overrides on-disk detection."""
+        bob = get_integration("bob")
+        (tmp_path / ".bob" / "skills" / "speckit-plan").mkdir(parents=True)
+        assert (
+            bob.is_skills_mode({"legacy_commands": True}, project_root=tmp_path)
+            is False
+        )
+
+    def test_explicit_skills_flag_forces_skills_over_legacy_disk_layout(
+        self, tmp_path
+    ):
+        """Regression (review #3415, 4724160183, comment 1).
+
+        ``--skills`` is the supported migration / opt-in: it must force skills
+        mode even when a managed legacy ``.bob/commands`` layout is on disk
+        (which otherwise auto-detects to legacy). This gives
+        ``integration upgrade bob --integration-options="--skills"`` a path out
+        of legacy mode instead of being trapped by disk detection.
+        """
+        bob = get_integration("bob")
+        cmds = tmp_path / ".bob" / "commands"
+        cmds.mkdir(parents=True)
+        (cmds / "speckit.plan.md").write_text("# plan", encoding="utf-8")
+        assert bob.is_skills_mode({"skills": True}, project_root=tmp_path) is True
+        assert (
+            bob.effective_invoke_separator({"skills": True}, project_root=tmp_path)
+            == "-"
+        )
+
+    def test_skills_and_legacy_flags_are_mutually_exclusive(self):
+        """Passing both ``--skills`` and ``--legacy-commands`` exits cleanly."""
+        import typer
+
+        bob = get_integration("bob")
+        with pytest.raises(typer.Exit):
+            bob.is_skills_mode({"skills": True, "legacy_commands": True})
+
+    def test_effective_invoke_separator_tracks_mode(self):
+        bob = get_integration("bob")
+        assert bob.effective_invoke_separator(None) == "-"
+        assert bob.effective_invoke_separator({"legacy_commands": True}) == "."
+        assert bob.effective_invoke_separator({"skills": True}) == "-"
+
+    def test_invoke_separator_for_mode_tracks_persisted_state(self):
+        """Registration paths resolve the separator from persisted ai_skills."""
+        bob = get_integration("bob")
+        assert bob.invoke_separator_for_mode(True) == "-"
+        assert bob.invoke_separator_for_mode(False) == "."
+
+    def test_no_skills_mode_method_leaks(self):
+        """The old callable _skills_mode method must be gone; consumers use the hook."""
+        bob = get_integration("bob")
+        assert not callable(getattr(bob, "_skills_mode", None))
+
+
+class TestBobDefaultSkillsMode:
+    """Default mode: .bob/skills/speckit-<name>/SKILL.md layout."""
+
+    def test_setup_creates_skill_files(self, tmp_path):
+        bob = get_integration("bob")
+        m = IntegrationManifest("bob", tmp_path)
+        created = bob.setup(tmp_path, m)
+        assert len(created) > 0
+        for f in created:
+            assert f.exists()
+            assert f.name == "SKILL.md"
+            assert f.parent.name.startswith("speckit-")
+
+    def test_setup_writes_to_correct_directory(self, tmp_path):
+        bob = get_integration("bob")
+        m = IntegrationManifest("bob", tmp_path)
+        bob.setup(tmp_path, m)
+        skills_dir = tmp_path / ".bob" / "skills"
+        assert skills_dir.is_dir()
+
+    def test_setup_does_not_warn(self, tmp_path):
+        bob = get_integration("bob")
+        m = IntegrationManifest("bob", tmp_path)
+        with warnings.catch_warnings(record=True) as caught:
+            warnings.simplefilter("always")
+            bob.setup(tmp_path, m)
+        assert not any(
+            "legacy" in str(item.message).lower() for item in caught
+        )
+
+    def test_setup_no_commands_dir(self, tmp_path):
+        bob = get_integration("bob")
+        m = IntegrationManifest("bob", tmp_path)
+        bob.setup(tmp_path, m)
+        assert not (tmp_path / ".bob" / "commands").exists()
+
+    def test_skill_directory_structure(self, tmp_path):
+        """Each command produces speckit-<name>/SKILL.md."""
+        bob = get_integration("bob")
+        m = IntegrationManifest("bob", tmp_path)
+        created = bob.setup(tmp_path, m)
+
+        expected_commands = {
+            "analyze", "clarify", "constitution", "converge", "implement",
+            "plan", "checklist", "specify", "tasks", "taskstoissues",
+        }
+        actual_commands = {f.parent.name.removeprefix("speckit-") for f in created}
+        assert actual_commands == expected_commands
+
+    def test_skill_frontmatter_structure(self, tmp_path):
+        """SKILL.md must have name, description, compatibility, metadata."""
+        bob = get_integration("bob")
+        m = IntegrationManifest("bob", tmp_path)
+        created = bob.setup(tmp_path, m)
+        for f in created:
+            content = f.read_text(encoding="utf-8")
+            assert content.startswith("---\n"), f"{f} missing frontmatter"
+            parts = content.split("---", 2)
+            fm = yaml.safe_load(parts[1])
+            assert "name" in fm
+            assert "description" in fm
+            assert "compatibility" in fm
+            assert "metadata" in fm
+            assert fm["metadata"]["author"] == "github-spec-kit"
+
+    def test_templates_are_processed(self, tmp_path):
+        bob = get_integration("bob")
+        m = IntegrationManifest("bob", tmp_path)
+        created = bob.setup(tmp_path, m)
+        for f in created:
+            content = f.read_text(encoding="utf-8")
+            assert "{SCRIPT}" not in content, f"{f.name} has unprocessed {{SCRIPT}}"
+            assert "__AGENT__" not in content, f"{f.name} has unprocessed __AGENT__"
+            assert "{ARGS}" not in content, f"{f.name} has unprocessed {{ARGS}}"
+            assert "__SPECKIT_COMMAND_" not in content, f"{f.name} has unprocessed __SPECKIT_COMMAND_*__"
+
+    def test_command_refs_use_hyphen_separator(self, tmp_path):
+        """Default skills layout must use /speckit-<name>, not /speckit.<name>."""
+        bob = get_integration("bob")
+        m = IntegrationManifest("bob", tmp_path)
+        created = bob.setup(tmp_path, m)
+        for f in created:
+            content = f.read_text(encoding="utf-8")
+            assert "/speckit." not in content, (
+                f"{f.name} contains dot-notation /speckit. reference; "
+                "skills must use /speckit-<name>"
+            )
+
+    def test_all_files_tracked_in_manifest(self, tmp_path):
+        bob = get_integration("bob")
+        m = IntegrationManifest("bob", tmp_path)
+        created = bob.setup(tmp_path, m)
+        for f in created:
+            rel = f.resolve().relative_to(tmp_path.resolve()).as_posix()
+            assert rel in m.files, f"{rel} not tracked in manifest"
+
+    def test_install_uninstall_roundtrip(self, tmp_path):
+        bob = get_integration("bob")
+        m = IntegrationManifest("bob", tmp_path)
+        created = bob.install(tmp_path, m)
+        assert len(created) > 0
+        m.save()
+        for f in created:
+            assert f.exists()
+        removed, skipped = bob.uninstall(tmp_path, m)
+        assert len(removed) == len(created)
+        assert skipped == []
+
+
+class TestBobLegacyCommandsMode:
+    """Legacy opt-in mode: .bob/commands/speckit.<name>.md layout."""
+
+    def test_setup_legacy_creates_markdown_files(self, tmp_path):
+        from specify_cli.integrations.bob import BobIntegration
+        bob = BobIntegration()
+        m = IntegrationManifest("bob", tmp_path)
+        created = bob.setup(tmp_path, m, parsed_options={"legacy_commands": True})
+        assert len(created) > 0
+        for f in created:
+            assert f.exists()
+            assert f.suffix == ".md"
+            assert f.name.startswith("speckit.")
+            assert f.parent == tmp_path / ".bob" / "commands"
+
+    def test_setup_legacy_warns_deprecated(self, tmp_path):
+        from specify_cli.integrations.bob import BobIntegration
+        bob = BobIntegration()
+        m = IntegrationManifest("bob", tmp_path)
+        with pytest.warns(UserWarning, match="Bob legacy commands mode"):
+            bob.setup(tmp_path, m, parsed_options={"legacy_commands": True})
+
+    def test_setup_legacy_no_skills_dir(self, tmp_path):
+        from specify_cli.integrations.bob import BobIntegration
+        bob = BobIntegration()
+        m = IntegrationManifest("bob", tmp_path)
+        bob.setup(tmp_path, m, parsed_options={"legacy_commands": True})
+        assert not (tmp_path / ".bob" / "skills").exists()
+
+    def test_setup_legacy_templates_are_processed(self, tmp_path):
+        from specify_cli.integrations.bob import BobIntegration
+        bob = BobIntegration()
+        m = IntegrationManifest("bob", tmp_path)
+        bob.setup(tmp_path, m, parsed_options={"legacy_commands": True})
+        commands_dir = tmp_path / ".bob" / "commands"
+        for md_file in commands_dir.glob("speckit.*.md"):
+            content = md_file.read_text(encoding="utf-8")
+            assert "{SCRIPT}" not in content
+            assert "__AGENT__" not in content
+            assert "{ARGS}" not in content
+            assert "__SPECKIT_COMMAND_" not in content
+
+    def test_setup_legacy_all_files_tracked(self, tmp_path):
+        from specify_cli.integrations.bob import BobIntegration
+        bob = BobIntegration()
+        m = IntegrationManifest("bob", tmp_path)
+        created = bob.setup(tmp_path, m, parsed_options={"legacy_commands": True})
+        for f in created:
+            rel = f.resolve().relative_to(tmp_path.resolve()).as_posix()
+            assert rel in m.files, f"{rel} not tracked in manifest"
+
+    def test_setup_legacy_uninstall_roundtrip(self, tmp_path):
+        from specify_cli.integrations.bob import BobIntegration
+        bob = BobIntegration()
+        m = IntegrationManifest("bob", tmp_path)
+        created = bob.install(tmp_path, m, parsed_options={"legacy_commands": True})
+        assert len(created) > 0
+        m.save()
+        removed, skipped = bob.uninstall(tmp_path, m)
+        assert len(removed) == len(created)
+        assert skipped == []
+
+
+class TestBobInitFlowDefault:
+    """CLI init creates skills by default."""
+
+    def test_init_default_creates_skills(self, tmp_path):
+        from typer.testing import CliRunner
+        from specify_cli import app
+
+        target = tmp_path / "test-proj"
+        result = CliRunner().invoke(app, [
+            "init", str(target), "--integration", "bob",
+            "--ignore-agent-tools", "--script", "sh",
+        ])
+        assert result.exit_code == 0, f"init --integration bob failed: {result.output}"
+        assert (target / ".bob" / "skills" / "speckit-plan" / "SKILL.md").exists()
+        assert not (target / ".bob" / "commands").exists()
+
+    def test_init_default_complete_file_inventory_sh(self, tmp_path):
+        from typer.testing import CliRunner
+        from specify_cli import app
+
+        project = tmp_path / "inventory-sh-bob"
+        project.mkdir()
+        old_cwd = os.getcwd()
+        try:
+            os.chdir(project)
+            result = CliRunner().invoke(app, [
+                "init", "--here", "--integration", "bob", "--script", "sh",
+                "--ignore-agent-tools",
+            ], catch_exceptions=False)
+        finally:
+            os.chdir(old_cwd)
+        assert result.exit_code == 0, f"init failed: {result.output}"
+
+        commands = [
+            "analyze", "clarify", "constitution", "converge", "implement",
+            "plan", "checklist", "specify", "tasks", "taskstoissues",
+        ]
+        for cmd in commands:
+            assert (project / ".bob" / "skills" / f"speckit-{cmd}" / "SKILL.md").exists(), (
+                f"Missing .bob/skills/speckit-{cmd}/SKILL.md"
+            )
+
+
+class TestBobInitFlowLegacy:
+    """CLI init with --legacy-commands produces .bob/commands/*.md."""
+
+    def test_init_legacy_creates_commands(self, tmp_path):
+        from typer.testing import CliRunner
+        from specify_cli import app
+
+        target = tmp_path / "test-proj"
+        result = CliRunner().invoke(app, [
+            "init", str(target), "--integration", "bob",
+            "--integration-options", "--legacy-commands",
+            "--ignore-agent-tools", "--script", "sh",
+        ])
+        assert result.exit_code == 0, f"init --integration bob --legacy-commands failed: {result.output}"
+        assert (target / ".bob" / "commands" / "speckit.plan.md").exists()
+        assert not (target / ".bob" / "skills").exists()
+
+    def test_init_legacy_does_not_set_ai_skills(self, tmp_path):
+        """Legacy install must NOT write ai_skills=True to init-options.json.
+
+        Behavioral guard for the dual-mode contract: with --legacy-commands,
+        BobIntegration.is_skills_mode(parsed_options) returns False, so
+        _update_init_options_for_integration must not persist ai_skills=True.
+        (Regression origin: shared code previously probed a bound _skills_mode
+        method object, which is always truthy, and wrongly enabled skills for
+        legacy projects.)
+        """
+        from typer.testing import CliRunner
+        from specify_cli import app
+        from specify_cli import load_init_options
+
+        target = tmp_path / "test-proj"
+        result = CliRunner().invoke(app, [
+            "init", str(target), "--integration", "bob",
+            "--integration-options", "--legacy-commands",
+            "--ignore-agent-tools", "--script", "sh",
+        ])
+        assert result.exit_code == 0, f"init failed: {result.output}"
+        init_opts = load_init_options(target)
+        assert init_opts.get("ai_skills") is not True, (
+            "Legacy Bob project must not have ai_skills=True in init-options.json"
+        )
+
+
+class TestBobRegistrarConfig:
+    """Verify AGENT_CONFIGS["bob"] follows the Copilot pattern for extension registration."""
+
+    def test_registrar_config_uses_commands_layout(self):
+        """AGENT_CONFIGS["bob"] must use the legacy .md layout (not /SKILL.md).
+
+        This mirrors Copilot: the static registrar config targets the non-skills
+        format so that:
+        - skills_mode_active becomes True when ai_skills=True, preventing
+          extension registration from writing SKILL.md files into .bob/skills/
+          on projects that never asked for legacy files.
+        - legacy-mode projects receive extension .md files in .bob/commands/.
+        """
+        from specify_cli.agents import CommandRegistrar
+        registrar = CommandRegistrar()
+        bob_cfg = registrar.AGENT_CONFIGS.get("bob")
+        assert bob_cfg is not None, "bob must be in AGENT_CONFIGS"
+        assert bob_cfg["extension"] == ".md", (
+            "AGENT_CONFIGS['bob']['extension'] must be '.md' so that "
+            "skills_mode_active=True suppresses extension registration on "
+            "skills-mode projects (mirrors the Copilot pattern)"
+        )
+        assert bob_cfg["dir"] == ".bob/commands"
+
+    def test_skills_mode_project_extension_registration_skipped(self, tmp_path):
+        """Extension registrar skips Bob on skills-mode projects (no .bob/commands dir)."""
+        from specify_cli.agents import CommandRegistrar
+        # Simulate a skills-mode Bob project: .bob/skills exists, .bob/commands does not
+        (tmp_path / ".bob" / "skills").mkdir(parents=True)
+
+        registrar = CommandRegistrar()
+        results = registrar.register_commands_for_all_agents(
+            commands=[{"name": "speckit.test-cmd", "file": "test.md"}],
+            source_id="test",
+            source_dir=tmp_path,
+            project_root=tmp_path,
+        )
+        # Bob must not appear in results — .bob/commands doesn't exist
+        assert "bob" not in results
+
+    def test_legacy_mode_project_extension_registration_runs(self, tmp_path):
+        """Extension registrar writes to .bob/commands/ for legacy-mode projects."""
+        import textwrap
+        from specify_cli.agents import CommandRegistrar
+
+        # Simulate a legacy-mode Bob project: .bob/commands exists, .bob/skills does not
+        commands_dir = tmp_path / ".bob" / "commands"
+        commands_dir.mkdir(parents=True)
+
+        # Provide a minimal command source file
+        cmd_file = tmp_path / "test.md"
+        cmd_file.write_text(
+            textwrap.dedent("""\
+                ---
+                description: "Test command"
+                ---
+                Test body.
+            """),
+            encoding="utf-8",
+        )
+
+        registrar = CommandRegistrar()
+        results = registrar.register_commands_for_all_agents(
+            commands=[{"name": "speckit.test-cmd", "file": "test.md"}],
+            source_id="test",
+            source_dir=tmp_path,
+            project_root=tmp_path,
+        )
+        assert "bob" in results, "bob must appear in results for legacy-mode project"
+        registered_file = commands_dir / "speckit.test-cmd.md"
+        assert registered_file.exists(), f"Expected {registered_file} to be written"
+
+    def test_legacy_extension_command_refs_use_dot_separator(self, tmp_path):
+        """Regression (review #3415): legacy .bob/commands/ extension commands must
+        render Bob 1.x ``/speckit.<cmd>`` refs, not the skills-layout ``/speckit-<cmd>``.
+
+        The single static AGENT_CONFIGS["bob"]["invoke_separator"] is "-" (the
+        default skills layout); register_commands must instead resolve the
+        separator from the project's persisted mode via
+        BobIntegration.invoke_separator_for_mode(False) -> ".".
+        """
+        import textwrap
+        from specify_cli.agents import CommandRegistrar
+
+        # Legacy-mode project: .bob/commands exists, ai_skills is NOT set.
+        commands_dir = tmp_path / ".bob" / "commands"
+        commands_dir.mkdir(parents=True)
+        cmd_file = tmp_path / "test.md"
+        cmd_file.write_text(
+            textwrap.dedent("""\
+                ---
+                description: "Test command"
+                ---
+                See __SPECKIT_COMMAND_SPECIFY__ for details.
+            """),
+            encoding="utf-8",
+        )
+
+        registrar = CommandRegistrar()
+        registrar.register_commands_for_all_agents(
+            commands=[{"name": "speckit.test-cmd", "file": "test.md"}],
+            source_id="test",
+            source_dir=tmp_path,
+            project_root=tmp_path,
+        )
+        rendered = (commands_dir / "speckit.test-cmd.md").read_text(encoding="utf-8")
+        assert "/speckit.specify" in rendered, (
+            "legacy Bob extension commands must render /speckit.specify (dot)"
+        )
+        assert "/speckit-specify" not in rendered
+
+
+class TestBobUseFlowPreservesLegacyLayout:
+    """Regression (review #3415): re-activating an existing Bob 1.x project
+    must not silently migrate it to the skills layout.
+    """
+
+    def test_update_init_options_preserves_legacy_commands_project(self, tmp_path):
+        """``use``/``switch``/``upgrade`` on a ``.bob/commands``-only project
+        (no stored ``legacy_commands``) must not write ``ai_skills=True``.
+        """
+        from specify_cli.integrations._helpers import (
+            _update_init_options_for_integration,
+        )
+        from specify_cli import load_init_options
+
+        # Existing Bob 1.x project: legacy commands dir on disk, no ai_skills.
+        cmds = tmp_path / ".bob" / "commands"
+        cmds.mkdir(parents=True)
+        (cmds / "speckit.plan.md").write_text("# plan", encoding="utf-8")
+        bob = get_integration("bob")
+
+        # Simulate the use/switch path: no parsed options were stored.
+        _update_init_options_for_integration(tmp_path, bob, parsed_options=None)
+
+        opts = load_init_options(tmp_path)
+        assert opts.get("ai") == "bob"
+        assert opts.get("ai_skills") is not True, (
+            "an existing .bob/commands project must stay legacy on re-activation"
+        )
+
+    def test_update_init_options_keeps_skills_project_as_skills(self, tmp_path):
+        """A ``.bob/skills`` project stays skills on re-activation."""
+        from specify_cli.integrations._helpers import (
+            _update_init_options_for_integration,
+        )
+        from specify_cli import load_init_options
+
+        (tmp_path / ".bob" / "skills" / "speckit-plan").mkdir(parents=True)
+        bob = get_integration("bob")
+
+        _update_init_options_for_integration(tmp_path, bob, parsed_options=None)
+
+        opts = load_init_options(tmp_path)
+        assert opts.get("ai_skills") is True
+
+    def test_with_integration_setting_stores_dot_separator_for_legacy(self, tmp_path):
+        """Regression (review #3415): shared-infra refresh on the use/switch
+        path resolves the command-ref separator *before* init-options are
+        rewritten, via ``effective_invoke_separator``.  For an existing
+        ``.bob/commands`` project with no stored options this must resolve to
+        ``"."`` (project-aware), not the skills-layout ``"-"``; otherwise core
+        command references get rewritten to ``/speckit-*``.
+        """
+        from specify_cli.integration_runtime import with_integration_setting
+
+        cmds = tmp_path / ".bob" / "commands"
+        cmds.mkdir(parents=True)
+        (cmds / "speckit.plan.md").write_text("# plan", encoding="utf-8")
+        bob = get_integration("bob")
+
+        # Simulate the use/switch path: no parsed options stored.
+        settings = with_integration_setting(
+            {}, "bob", bob, parsed_options=None, project_root=tmp_path
+        )
+        assert settings["bob"]["invoke_separator"] == ".", (
+            "legacy .bob/commands project must persist the dot separator so "
+            "shared templates render Bob 1.x /speckit.<cmd> references"
+        )
+
+    def test_use_force_keeps_legacy_command_refs_in_shared_templates(self, tmp_path):
+        """End-to-end (review #3415): ``integration use bob --force`` on an
+        existing Bob 1.x project (legacy layout on disk, stored options
+        stripped as a pre-PR install would be) must re-render shared templates
+        with ``/speckit.<cmd>`` (dot), not ``/speckit-<cmd>``.
+        """
+        import json
+        from typer.testing import CliRunner
+        from specify_cli import app
+
+        # Create a real legacy Bob project (renders shared templates).
+        target = tmp_path / "proj"
+        runner = CliRunner()
+        result = runner.invoke(app, [
+            "init", str(target), "--integration", "bob",
+            "--integration-options", "--legacy-commands",
+            "--ignore-agent-tools", "--script", "sh",
+        ])
+        assert result.exit_code == 0, f"init failed: {result.output}"
+
+        template = target / ".specify" / "templates" / "plan-template.md"
+        assert template.is_file(), "expected a rendered shared plan template"
+        assert "/speckit.plan" in template.read_text(encoding="utf-8")
+
+        # Simulate a pre-PR Bob 1.x install: no stored options/separator.
+        integ_json = target / ".specify" / "integration.json"
+        data = json.loads(integ_json.read_text(encoding="utf-8"))
+        bob_settings = data["integration_settings"]["bob"]
+        for stale in ("raw_options", "parsed_options", "invoke_separator"):
+            bob_settings.pop(stale, None)
+        integ_json.write_text(json.dumps(data, indent=2), encoding="utf-8")
+
+        # Re-activate with --force so shared templates are re-rendered.
+        import os
+        old_cwd = os.getcwd()
+        try:
+            os.chdir(target)
+            result = runner.invoke(
+                app, ["integration", "use", "bob", "--force"]
+            )
+        finally:
+            os.chdir(old_cwd)
+        assert result.exit_code == 0, f"use failed: {result.output}"
+
+        rendered = template.read_text(encoding="utf-8")
+        assert "/speckit.plan" in rendered, (
+            "legacy Bob project must keep /speckit.plan (dot) after refresh"
+        )
+        assert "/speckit-plan" not in rendered, (
+            "shared templates must not be rewritten to the skills /speckit-plan"
+        )
+        # And the persisted separator must reflect the legacy layout.
+        data = json.loads(integ_json.read_text(encoding="utf-8"))
+        assert data["integration_settings"]["bob"].get("invoke_separator") == "."
+
+
+class TestBobCommandRefScopedToActiveAgent:
+    """Regression (review #3415, 4716424313).
+
+    ``CommandRegistrar.register_commands`` runs once per detected agent, but the
+    persisted ``ai_skills`` flag describes only the *active* integration
+    (``opts["ai"]``).  When another agent (e.g. Copilot) is active in skills
+    mode while a legacy ``.bob/commands`` layout is also present, Bob's command
+    references must still render with the ``.`` separator (Bob 1.x
+    ``/speckit.<cmd>``) rather than inheriting Copilot's ``ai_skills=True`` and
+    rendering ``/speckit-<cmd>``.
+    """
+
+    def _write_command_ref_ext(self, source_dir):
+        source_dir.mkdir(parents=True, exist_ok=True)
+        cmd = source_dir / "run.md"
+        cmd.write_text(
+            "---\ndescription: Run\n---\n\nUse __SPECKIT_COMMAND_PLAN__ first.\n",
+            encoding="utf-8",
+        )
+        return [{"name": "speckit.ext.run", "file": "run.md"}]
+
+    def test_legacy_bob_ref_not_rewritten_when_other_agent_active_in_skills(
+        self, tmp_path
+    ):
+        from specify_cli._init_options import save_init_options
+        from specify_cli.agents import CommandRegistrar
+
+        # Legacy Bob layout on disk; skills layout absent.
+        (tmp_path / ".bob" / "commands").mkdir(parents=True)
+        # A different agent (Copilot) is the active integration, in skills mode.
+        save_init_options(tmp_path, {"ai": "copilot", "ai_skills": True})
+
+        source_dir = tmp_path / "ext-src"
+        commands = self._write_command_ref_ext(source_dir)
+
+        registrar = CommandRegistrar()
+        registered = registrar.register_commands(
+            "bob", commands, "ext", source_dir, tmp_path,
+        )
+        assert "speckit.ext.run" in registered
+
+        written = list((tmp_path / ".bob" / "commands").glob("*.md"))
+        assert written, "expected a rendered Bob command file"
+        content = written[0].read_text(encoding="utf-8")
+        assert "__SPECKIT_COMMAND_PLAN__" not in content
+        assert "/speckit.plan" in content, (
+            "legacy Bob command refs must use the dot separator even when "
+            "another agent is active in skills mode"
+        )
+        assert "/speckit-plan" not in content
+
+    def test_active_bob_skills_command_output_uses_dot(self, tmp_path):
+        """Regression (review #3415, 4724160183, comment 2).
+
+        The separator must match the *output layout* the registrar writes, not
+        the project's persisted ``ai_skills`` flag.  Even when Bob itself is the
+        active agent in skills mode, a ``.bob/commands/*.md`` file is a
+        command-layout artifact and must render Bob 1.x ``/speckit.<cmd>``.
+        Rendering ``/speckit-<cmd>`` into a command file (as the old
+        ``ai_skills``-driven active-agent branch did) produced an invocation the
+        command layout can't resolve.  Bob skills are written via its own
+        skills path, so ``register_commands`` only ever emits command-layout
+        files for Bob.
+        """
+        from specify_cli._init_options import save_init_options
+        from specify_cli.agents import CommandRegistrar
+
+        (tmp_path / ".bob" / "skills").mkdir(parents=True)
+        save_init_options(tmp_path, {"ai": "bob", "ai_skills": True})
+
+        source_dir = tmp_path / "ext-src"
+        commands = self._write_command_ref_ext(source_dir)
+
+        registrar = CommandRegistrar()
+        registrar.register_commands("bob", commands, "ext", source_dir, tmp_path)
+
+        written = list((tmp_path / ".bob" / "commands").glob("*.md"))
+        assert written, "expected a rendered Bob command file"
+        content = written[0].read_text(encoding="utf-8")
+        assert "__SPECKIT_COMMAND_PLAN__" not in content
+        assert "/speckit.plan" in content, (
+            "a .bob/commands/*.md command-layout file must use the dot "
+            "separator even when Bob is the active agent in skills mode"
+        )
+        assert "/speckit-plan" not in content
+
+    def test_inactive_bob_command_output_uses_dot_even_with_skills_dir(
+        self, tmp_path
+    ):
+        """Regression (review #3415, 4723246468): for an inactive Bob install
+        the registrar's separator must match the layout it is actually writing
+        (``.bob/commands/*.md`` — command layout), not on-disk sibling dirs.
+        Even when a ``.bob/skills/`` directory (with managed ``speckit-*``
+        skills) coexists, command-layout files must keep ``/speckit.<cmd>``.
+        """
+        from specify_cli._init_options import save_init_options
+        from specify_cli.agents import CommandRegistrar
+
+        # Both layouts on disk; the active agent is something else entirely.
+        (tmp_path / ".bob" / "commands").mkdir(parents=True)
+        (tmp_path / ".bob" / "skills" / "speckit-plan").mkdir(parents=True)
+        save_init_options(tmp_path, {"ai": "claude", "ai_skills": True})
+
+        source_dir = tmp_path / "ext-src"
+        commands = self._write_command_ref_ext(source_dir)
+
+        registrar = CommandRegistrar()
+        registrar.register_commands("bob", commands, "ext", source_dir, tmp_path)
+
+        written = list((tmp_path / ".bob" / "commands").glob("*.md"))
+        assert written, "expected a rendered Bob command file"
+        content = written[0].read_text(encoding="utf-8")
+        assert "__SPECKIT_COMMAND_PLAN__" not in content
+        assert "/speckit.plan" in content, (
+            "Bob command-layout output must use the dot separator regardless "
+            "of a coexisting .bob/skills directory"
+        )
+        assert "/speckit-plan" not in content
+
+
+class TestBobSetupPreservesLegacyOnUpgrade:
+    """Regression (review #3415, 4723782860, comment 1).
+
+    ``setup()`` must apply the same managed-artifact detection as ``use`` so
+    that ``integration upgrade bob`` on a Bob 1.x install (managed
+    ``.bob/commands/speckit.*.md`` on disk, no stored options) preserves the
+    command layout instead of silently generating skills and stale-deleting
+    the legacy commands.
+    """
+
+    def test_setup_without_options_preserves_existing_command_layout(
+        self, tmp_path
+    ):
+        from specify_cli.integrations.bob import BobIntegration
+
+        # Pre-existing Bob 1.x install: managed command files, no options.
+        cmds = tmp_path / ".bob" / "commands"
+        cmds.mkdir(parents=True)
+        (cmds / "speckit.plan.md").write_text("# plan", encoding="utf-8")
+
+        bob = BobIntegration()
+        m = IntegrationManifest("bob", tmp_path)
+        with pytest.warns(UserWarning, match="Bob legacy commands mode"):
+            created = bob.setup(tmp_path, m, parsed_options=None)
+
+        # Command layout regenerated; no skills layout introduced.
+        assert not (tmp_path / ".bob" / "skills").exists(), (
+            "upgrade must not migrate an existing legacy Bob project to skills"
+        )
+        assert created, "expected command files to be regenerated"
+        for f in created:
+            assert f.parent == tmp_path / ".bob" / "commands"
+            assert f.suffix == ".md"
+
+    def test_setup_fresh_project_still_defaults_to_skills(self, tmp_path):
+        """A fresh project (no managed artifacts) still defaults to skills."""
+        from specify_cli.integrations.bob import BobIntegration
+
+        bob = BobIntegration()
+        m = IntegrationManifest("bob", tmp_path)
+        created = bob.setup(tmp_path, m, parsed_options=None)
+
+        assert (tmp_path / ".bob" / "skills").is_dir()
+        assert not (tmp_path / ".bob" / "commands").exists()
+        assert created
+
+    def test_setup_with_skills_flag_migrates_legacy_to_skills(self, tmp_path):
+        """Review #3415, 4724160183, comment 1: ``--skills`` on an existing
+        legacy install forces the skills layout (the migration opt-in), instead
+        of preserving the auto-detected legacy layout. ``setup()`` scaffolds the
+        skills layout; the ``integration upgrade`` stale-file pass removes the
+        old command files.
+        """
+        from specify_cli.integrations.bob import BobIntegration
+
+        # Pre-existing Bob 1.x install on disk.
+        cmds = tmp_path / ".bob" / "commands"
+        cmds.mkdir(parents=True)
+        (cmds / "speckit.plan.md").write_text("# plan", encoding="utf-8")
+
+        bob = BobIntegration()
+        m = IntegrationManifest("bob", tmp_path)
+        # No deprecation warning — the user opted into skills, not legacy.
+        with warnings.catch_warnings():
+            warnings.simplefilter("error", UserWarning)
+            created = bob.setup(tmp_path, m, parsed_options={"skills": True})
+
+        assert (tmp_path / ".bob" / "skills").is_dir(), (
+            "--skills must force the skills layout even when a legacy commands "
+            "layout is already on disk"
+        )
+        assert created
+        for f in created:
+            assert f.name == "SKILL.md"
+            assert f.parent.name.startswith("speckit-")
+
+
+class TestBobPostProcessSkillContent:
+    """Regression (review #3415, 4723782860, comment 2).
+
+    Preset/extension skill generators call ``post_process_skill_content`` on
+    the *registered* ``BobIntegration`` instance. Core Bob skills are
+    intent-activated and intentionally omit the shared slash-command hook note,
+    so the registered class must expose the same no-op the skills helper does
+    (not inherit a note-injecting default) to keep every skill path consistent.
+    """
+
+    def test_registered_bob_has_post_process_hook(self):
+        bob = get_integration("bob")
+        assert hasattr(bob, "post_process_skill_content")
+
+    def test_post_process_is_noop_no_hook_note_injected(self):
+        bob = get_integration("bob")
+        sample = (
+            "---\nname: speckit-plan\n---\n\n"
+            "Run /speckit.plan then /speckit.tasks.\n"
+        )
+        assert bob.post_process_skill_content(sample) == sample
+
+    def test_post_process_matches_skills_helper(self):
+        from specify_cli.integrations.bob import _BobSkillsHelper
+
+        bob = get_integration("bob")
+        sample = "---\nname: speckit-analyze\n---\n\nSome body with /speckit.plan.\n"
+        assert (
+            bob.post_process_skill_content(sample)
+            == _BobSkillsHelper().post_process_skill_content(sample)
+        )

tests/integrations/test_integration_catalog.py

@@ -13,9 +13,34 @@ from specify_cli.integrations.catalog import (
     IntegrationDescriptor,
     IntegrationDescriptorError,
     IntegrationValidationError,
+    _catalog_shape_error,
 )
 
 
+class TestCatalogShapeValidator:
+    """The shared shape validator used by BOTH the fresh-fetch and cache-read
+    paths, so a poisoned/older cache can't bypass the format contract the fresh
+    fetch enforces (dict + 'schema_version' + dict 'integrations')."""
+
+    def test_valid_payload_returns_none(self):
+        assert _catalog_shape_error({"schema_version": "1.0", "integrations": {}}) is None
+
+    def test_missing_schema_version_is_rejected(self):
+        # The exact bypass the two paths used to disagree on: a dict with a dict
+        # 'integrations' but no 'schema_version'.
+        assert _catalog_shape_error({"integrations": {}}) is not None
+
+    def test_missing_integrations_is_rejected(self):
+        assert _catalog_shape_error({"schema_version": "1.0"}) is not None
+
+    def test_non_dict_integrations_is_rejected(self):
+        assert _catalog_shape_error({"schema_version": "1.0", "integrations": []}) is not None
+
+    @pytest.mark.parametrize("payload", [[], "x", 5, None])
+    def test_non_dict_payload_is_rejected(self, payload):
+        assert _catalog_shape_error(payload) is not None
+
+
 # ---------------------------------------------------------------------------
 # IntegrationCatalogEntry
 # ---------------------------------------------------------------------------
@@ -251,6 +276,48 @@ class TestCatalogFetch:
         ids = [r["id"] for r in results]
         assert "acme-coder" in ids
 
+    def test_poisoned_cache_shape_is_dropped_and_refetched(self, tmp_path, monkeypatch):
+        """A fresh-but-mis-shaped cache (e.g. integrations as a list) must be
+        dropped and refetched, not returned — otherwise it later crashes on
+        .items(). The cache path must clear the same shape checks as a fresh
+        fetch."""
+        monkeypatch.setenv("HOME", str(tmp_path))
+        monkeypatch.setenv("USERPROFILE", str(tmp_path))
+        monkeypatch.delenv("SPECKIT_INTEGRATION_CATALOG_URL", raising=False)
+        (tmp_path / ".specify").mkdir()
+        cat = IntegrationCatalog(tmp_path)
+
+        catalog = {
+            "schema_version": "1.0",
+            "updated_at": "2026-01-01T00:00:00Z",
+            "integrations": {
+                "acme-coder": {
+                    "id": "acme-coder", "name": "Acme Coder", "version": "2.0.0",
+                    "description": "Community integration", "author": "acme-org",
+                    "tags": ["cli"],
+                },
+            },
+        }
+        self._patch_urlopen(monkeypatch, catalog)
+        cat.search()  # populate the cache legitimately
+
+        # Poison the cached payload (integrations as a list), keeping the fresh
+        # metadata so the age check passes and the cache branch is taken.
+        cache_dir = tmp_path / ".specify" / "integrations" / ".cache"
+        data_files = [
+            f for f in cache_dir.glob("catalog-*.json")
+            if not f.name.endswith("-metadata.json")
+        ]
+        assert data_files, "cache was not populated"
+        data_files[0].write_text(
+            json.dumps({"schema_version": "1.0", "integrations": []}),
+            encoding="utf-8",
+        )
+
+        # The poisoned cache is dropped and the (valid) source is refetched.
+        results = cat.search()
+        assert "acme-coder" in [r["id"] for r in results]
+
     def test_search_by_tag(self, tmp_path, monkeypatch):
         monkeypatch.setenv("HOME", str(tmp_path))
         monkeypatch.setenv("USERPROFILE", str(tmp_path))
@@ -922,6 +989,57 @@ class TestCatalogSourceManagement:
         assert str(cfg_path) in message
         assert "expected a mapping" in message
 
+    def test_add_catalog_rejects_inf_priority_in_existing_entry(
+        self, tmp_path, monkeypatch
+    ):
+        # ``priority: .inf`` loads as float('inf'); int() on it raises
+        # OverflowError, which used to escape the IntegrationValidationError
+        # contract as a raw traceback (github/spec-kit#3526 fixed the sibling
+        # workflow/step loaders the same way).
+        self._isolate(tmp_path, monkeypatch)
+        cfg_path = tmp_path / ".specify" / "integration-catalogs.yml"
+        cfg_path.write_text(
+            yaml.dump(
+                {
+                    "catalogs": [
+                        {
+                            "url": "https://a.example.com/catalog.json",
+                            "priority": float("inf"),
+                        }
+                    ]
+                }
+            ),
+            encoding="utf-8",
+        )
+        cat = IntegrationCatalog(tmp_path)
+        with pytest.raises(
+            IntegrationValidationError, match="must be an integer"
+        ):
+            cat.add_catalog("https://new.example.com/catalog.json")
+
+    def test_remove_catalog_tolerates_inf_priority(self, tmp_path, monkeypatch):
+        # Building the remove display order must not crash on a ``priority:
+        # .inf`` entry; it falls back to positional order like the other
+        # non-integer priorities do.
+        self._isolate(tmp_path, monkeypatch)
+        cfg_path = tmp_path / ".specify" / "integration-catalogs.yml"
+        cfg_path.write_text(
+            yaml.dump(
+                {
+                    "catalogs": [
+                        {
+                            "url": "https://a.example.com/catalog.json",
+                            "priority": float("inf"),
+                        },
+                        {"url": "https://b.example.com/catalog.json", "priority": 2},
+                    ]
+                }
+            ),
+            encoding="utf-8",
+        )
+        cat = IntegrationCatalog(tmp_path)
+        cat.remove_catalog(0)  # must not raise OverflowError
+
     def test_add_catalog_skips_blank_url_entries(self, tmp_path, monkeypatch):
         self._isolate(tmp_path, monkeypatch)
         cfg_path = tmp_path / ".specify" / "integration-catalogs.yml"

tests/integrations/test_integration_copilot.py

@@ -575,6 +575,17 @@ class TestCopilotSkillsMode:
         assert copilot.effective_invoke_separator({"skills": True}) == "-"
         assert copilot.effective_invoke_separator({"skills": False}) == "."
 
+    def test_invoke_separator_for_mode_tracks_persisted_state(self):
+        """Regression (review #3415): registration paths (preset/extension
+        command refs) must resolve the separator from the persisted ai_skills
+        state. A Copilot skills project renders ``/speckit-<cmd>`` (hyphen),
+        matching ``build_command_invocation``; the default markdown layout
+        renders ``/speckit.<cmd>`` (dot).
+        """
+        copilot = self._make_copilot()
+        assert copilot.invoke_separator_for_mode(True) == "-"
+        assert copilot.invoke_separator_for_mode(False) == "."
+
     def test_skill_body_has_content(self, tmp_path):
         """Each SKILL.md body should contain template content."""
         copilot = self._make_copilot()

tests/integrations/test_integration_droid.py

@@ -0,0 +1,262 @@
+"""Tests for DroidIntegration (Factory Droid CLI)."""
+
+from urllib.parse import urlparse
+
+import pytest
+
+from specify_cli.integrations import get_integration
+from specify_cli.integrations.droid import DroidIntegration
+from specify_cli.integrations.manifest import IntegrationManifest
+
+from .test_integration_base_skills import SkillsIntegrationTests
+
+
+class TestDroidIntegration(SkillsIntegrationTests):
+    KEY = "droid"
+    FOLDER = ".factory/"
+    COMMANDS_SUBDIR = "skills"
+    REGISTRAR_DIR = ".factory/skills"
+
+    def test_options_include_skills_flag(self):
+        """Not applicable — Droid only supports the skills layout."""
+        pytest.skip("Droid is always skills-based and does not expose a --skills option")
+
+    def test_options_do_not_include_skills_flag(self):
+        """Droid is always skills-based; no --skills option is exposed."""
+        i = get_integration(self.KEY)
+        assert i is not None
+        opts = i.options()
+        skills_opts = [o for o in opts if o.name == "--skills"]
+        assert len(skills_opts) == 0, (
+            "Droid is always skills-based and should not expose a --skills option"
+        )
+
+    def test_requires_cli_is_true(self):
+        """Droid is a CLI tool; requires_cli must be True."""
+        i = get_integration(self.KEY)
+        assert i is not None
+        assert i.config["requires_cli"] is True
+        assert i.config["name"] == "Factory Droid"
+
+    def test_multi_install_safe_is_true(self):
+        """Droid uses an isolated .factory/ root — safe to install alongside others."""
+        i = get_integration(self.KEY)
+        assert i.multi_install_safe is True
+
+    def test_install_url_points_to_factory(self):
+        i = get_integration(self.KEY)
+        url = i.config.get("install_url")
+        assert url is not None
+        host = (urlparse(url).hostname or "").lower()
+        assert host == "factory.ai" or host.endswith(".factory.ai"), (
+            f"install_url must point at the Factory domain, got: {url}"
+        )
+
+
+class TestDroidInitFlow:
+    """--integration droid creates expected files."""
+
+    def test_integration_droid_creates_skills(self, tmp_path):
+        """--integration droid should create skills under .factory/skills."""
+        from typer.testing import CliRunner
+
+        from specify_cli import app
+
+        runner = CliRunner()
+        target = tmp_path / "test-proj"
+        result = runner.invoke(
+            app,
+            [
+                "init",
+                str(target),
+                "--integration",
+                "droid",
+                "--ignore-agent-tools",
+                "--script",
+                "sh",
+            ],
+            catch_exceptions=False,
+        )
+
+        assert result.exit_code == 0, f"init --integration droid failed: {result.output}"
+        assert (target / ".factory" / "skills" / "speckit-plan" / "SKILL.md").exists()
+        assert (target / ".factory" / "skills" / "speckit-specify" / "SKILL.md").exists()
+
+
+class TestDroidBuildExecArgs:
+    """Droid non-interactive execution argument building."""
+
+    def test_default_argv_uses_exec_subcommand(self):
+        """Default argv: ``droid exec <prompt> --output-format json``.
+
+        No permission-bypass flag is auto-applied — operators who need it
+        must pass it through ``SPECKIT_INTEGRATION_DROID_EXTRA_ARGS``.
+        """
+        i = get_integration("droid")
+        args = i.build_exec_args("/speckit-specify some-feature")
+        assert args == [
+            "droid",
+            "exec",
+            "/speckit-specify some-feature",
+            "--output-format",
+            "json",
+        ]
+        assert "--skip-permissions-unsafe" not in args, (
+            "Spec Kit must not auto-apply --skip-permissions-unsafe; "
+            "it is a dangerous flag and operators must opt in explicitly"
+        )
+
+    def test_text_output_omits_format_flag(self):
+        i = get_integration("droid")
+        args = i.build_exec_args("/speckit-plan", output_json=False)
+        assert args == [
+            "droid",
+            "exec",
+            "/speckit-plan",
+        ]
+        assert "--skip-permissions-unsafe" not in args
+
+    def test_model_is_appended(self):
+        i = get_integration("droid")
+        args = i.build_exec_args(
+            "/speckit-specify", model="claude-opus-4-7", output_json=False
+        )
+        assert args == [
+            "droid",
+            "exec",
+            "/speckit-specify",
+            "--model",
+            "claude-opus-4-7",
+        ]
+        assert "--skip-permissions-unsafe" not in args
+
+    def test_extra_args_inserted_after_canonical_flags(self, monkeypatch):
+        """Operator-injected extra args land after Spec Kit's canonical
+        ``--model`` / ``--output-format`` flags so the canonical flags are
+        always present in argv regardless of operator override."""
+        from specify_cli.integrations import get_integration
+
+        i = get_integration("droid")
+        monkeypatch.setenv("SPECKIT_INTEGRATION_DROID_EXTRA_ARGS", "--foo bar")
+        args = i.build_exec_args(
+            "/speckit-plan", model="claude-sonnet", output_json=True
+        )
+
+        assert "--foo" in args
+        assert "bar" in args
+        assert args.index("bar") == args.index("--foo") + 1
+        # Extra args land AFTER the canonical flags so the canonical flags
+        # are always present in argv.
+        assert args.index("--model") < args.index("--foo")
+        assert args.index("--output-format") < args.index("--foo")
+        assert args[args.index("--model") + 1] == "claude-sonnet"
+        assert args[args.index("--output-format") + 1] == "json"
+
+    def test_executable_override(self, monkeypatch):
+        """``SPECKIT_INTEGRATION_DROID_EXECUTABLE`` overrides argv[0]."""
+        monkeypatch.setenv(
+            "SPECKIT_INTEGRATION_DROID_EXECUTABLE", "/custom/droid"
+        )
+        i = get_integration("droid")
+        args = i.build_exec_args("/speckit-plan", output_json=False)
+        assert args[0] == "/custom/droid"
+        # No dangerous permission-bypass flag should leak in via the override path.
+        assert "--skip-permissions-unsafe" not in args
+
+    def test_returns_none_when_requires_cli_is_false(self, monkeypatch):
+        """When ``requires_cli`` is False, ``build_exec_args`` returns None."""
+        i = get_integration("droid")
+        monkeypatch.setitem(i.config, "requires_cli", False)
+        assert i.build_exec_args("/speckit-plan") is None
+
+
+class TestDroidFrontmatter:
+    """Every generated SKILL.md must carry Droid-specific frontmatter flags."""
+
+    def test_skills_carry_user_invocable_true(self, tmp_path):
+        i = get_integration("droid")
+        m = IntegrationManifest("droid", tmp_path)
+        i.setup(tmp_path, m, script_type="sh")
+
+        skill_files = [
+            f
+            for f in (tmp_path / ".factory" / "skills").rglob("SKILL.md")
+        ]
+        assert skill_files, "expected at least one SKILL.md"
+        for f in skill_files:
+            content = f.read_text(encoding="utf-8")
+            assert "user-invocable: true" in content, (
+                f"{f} missing user-invocable: true"
+            )
+
+    def test_skills_carry_disable_model_invocation_false(self, tmp_path):
+        i = get_integration("droid")
+        m = IntegrationManifest("droid", tmp_path)
+        i.setup(tmp_path, m, script_type="sh")
+
+        skill_files = [
+            f
+            for f in (tmp_path / ".factory" / "skills").rglob("SKILL.md")
+        ]
+        assert skill_files, "expected at least one SKILL.md"
+        for f in skill_files:
+            content = f.read_text(encoding="utf-8")
+            assert "disable-model-invocation: false" in content, (
+                f"{f} missing disable-model-invocation: false"
+            )
+
+    def test_inject_frontmatter_flag_adds_key_when_absent(self):
+        """Fresh content (key absent) gets the flag injected on its own line."""
+        content = "---\nname: x\ndescription: y\n---\n\nBody.\n"
+        result = DroidIntegration._inject_frontmatter_flag(content, "user-invocable")
+        assert "user-invocable: true" in result
+        # The injected key must sit on its own line, not glued to the closing ---.
+        assert "\nuser-invocable: true\n---" in result, (
+            "Injected key must be on its own line, not fused to closing ---"
+        )
+
+    def test_inject_frontmatter_flag_injects_custom_value(self):
+        """The value parameter must be honored (used for disable-model-invocation: false)."""
+        content = "---\nname: x\n---\n\nBody.\n"
+        result = DroidIntegration._inject_frontmatter_flag(
+            content, "disable-model-invocation", "false"
+        )
+        assert "disable-model-invocation: false" in result
+
+    def test_inject_frontmatter_flag_no_trailing_newline(self):
+        """Regression for the frontmatter-fusion P2 bug.
+
+        When the closing ``---`` is the literal last line of the file with
+        no trailing newline, the injected key must still land on its own
+        line (not fused onto the closing delimiter). Previously this
+        produced ``user-invocable: true---``, an unparseable YAML line.
+        """
+        content = "---\nname: x\ndescription: y\n---"
+        result = DroidIntegration._inject_frontmatter_flag(content, "user-invocable")
+        assert "user-invocable: true" in result
+        # The injected key and the closing delimiter must NOT be fused.
+        assert "user-invocable: true---" not in result, (
+            "Injected key fused onto closing ---; no-trailing-newline regression"
+        )
+        # And the injected key must be on its own line.
+        assert "\nuser-invocable: true\n---" in result
+
+    def test_frontmatter_injection_is_idempotent(self):
+        """Running the post-processor twice must not duplicate the flag."""
+        content = "---\nname: x\n---\n\nBody.\n"
+        once = DroidIntegration._inject_frontmatter_flag(content, "user-invocable")
+        twice = DroidIntegration._inject_frontmatter_flag(once, "user-invocable")
+        assert once == twice, "Frontmatter injection must be idempotent"
+        # Belt-and-braces: the flag must appear exactly once.
+        assert once.count("user-invocable: true") == 1
+
+
+class TestDroidCommandInvocation:
+    """Skills agents use the hyphenated ``/speckit-<name>`` slash form."""
+
+    def test_build_command_invocation_uses_hyphenated_skill_name(self):
+        i = get_integration("droid")
+        assert i.build_command_invocation("speckit.plan", "feature-x") == (
+            "/speckit-plan feature-x"
+        )
+        assert i.build_command_invocation("plan") == "/speckit-plan"

tests/integrations/test_integration_forge.py

@@ -475,3 +475,39 @@ class TestForgeCommandRegistrar:
             "Found '/speckit.specify' (dot notation) in generated Forge git.feature command body. "
             "Forge requires hyphen notation for ZSH compatibility."
         )
+
+
+class TestForgeInitNextSteps:
+    """The post-init 'Next steps' panel must show hyphenated /speckit-<name>
+    commands for Forge, since Forge only registers the hyphenated form
+    (see the generated command-file tests above)."""
+
+    def test_init_next_steps_show_hyphenated_commands(self, tmp_path):
+        import os
+
+        from typer.testing import CliRunner
+
+        from specify_cli import app
+
+        project = tmp_path / "forge-nextsteps"
+        project.mkdir()
+        old_cwd = os.getcwd()
+        try:
+            os.chdir(project)
+            result = CliRunner().invoke(
+                app,
+                ["init", "--here", "--integration", "forge", "--ignore-agent-tools"],
+                catch_exceptions=False,
+            )
+        finally:
+            os.chdir(old_cwd)
+
+        assert result.exit_code == 0, f"init failed: {result.output}"
+        # Forge registers /speckit-<name>; the next-steps panel must match.
+        assert "/speckit-plan" in result.output, (
+            f"Expected /speckit-plan in next steps but got:\n{result.output}"
+        )
+        # Must NOT show the dotted /speckit.plan form Forge can't invoke.
+        assert "/speckit.plan" not in result.output, (
+            f"Should not show dotted /speckit.plan for Forge:\n{result.output}"
+        )

tests/integrations/test_integration_subcommand.py

@@ -15,6 +15,22 @@ from tests.conftest import strip_ansi
 runner = CliRunner()
 
 
+@pytest.mark.parametrize(
+    "args",
+    [
+        ["init", "--help"],
+        ["integration", "install", "--help"],
+        ["integration", "switch", "--help"],
+        ["integration", "upgrade", "--help"],
+    ],
+)
+def test_script_help_includes_python_variant(args):
+    result = runner.invoke(app, args)
+
+    assert result.exit_code == 0
+    assert "sh, ps, or py" in " ".join(strip_ansi(result.output).split())
+
+
 def _init_project(tmp_path, integration="copilot", integration_options=None):
     """Helper: init a spec-kit project with the given integration."""
     project = tmp_path / "proj"
@@ -2477,6 +2493,300 @@ class TestIntegrationUpgrade:
             f"found: {[f.name for f in core_remaining]}"
         )
 
+    def test_upgrade_bob_skills_migration_preserves_manifest(self, tmp_path):
+        """Regression (review #3415, 4724160183, comment 1).
+
+        ``integration upgrade bob --integration-options="--skills"`` migrates a
+        legacy Bob 1.x install (``.bob/commands/*.md``) to the skills layout
+        (``.bob/skills/speckit-*/SKILL.md``) and stale-removes the old command
+        files.  Because that stale-file pass shrinks the tracked set, the
+        upgrade's Phase 2 must NOT delete the freshly-saved ``bob.manifest.json``
+        — otherwise the migrated project is left untracked and un-upgradeable.
+        """
+        project = _init_project(
+            tmp_path, "bob", integration_options="--legacy-commands"
+        )
+
+        commands = project / ".bob" / "commands"
+        skills = project / ".bob" / "skills"
+        manifest_path = (
+            project / ".specify" / "integrations" / "bob.manifest.json"
+        )
+        assert commands.is_dir() and sorted(commands.glob("speckit.*.md"))
+        assert not skills.exists()
+        assert manifest_path.is_file()
+
+        result = _run_in_project(project, [
+            "integration", "upgrade", "bob",
+            "--integration-options", "--skills",
+            "--script", "sh", "--force",
+        ])
+        assert result.exit_code == 0, f"migration upgrade failed: {result.output}"
+
+        # Skills layout scaffolded; legacy core command files removed.
+        assert skills.is_dir(), ".bob/skills/ must exist after --skills migration"
+        assert sorted(skills.glob("speckit-*")), "expected migrated skill dirs"
+        core_commands = [
+            f for f in commands.glob("speckit.*.md")
+            if "agent-context" not in f.name
+        ] if commands.exists() else []
+        assert core_commands == [], (
+            f"legacy core command files should be removed, found: "
+            f"{[f.name for f in core_commands]}"
+        )
+
+        # The manifest must survive so the project stays tracked/upgradeable.
+        assert manifest_path.is_file(), (
+            "bob.manifest.json must survive a layout-shrinking migration"
+        )
+        reupgrade = _run_in_project(project, [
+            "integration", "upgrade", "bob", "--script", "sh", "--force",
+        ])
+        assert reupgrade.exit_code == 0, (
+            f"migrated project must remain upgradeable: {reupgrade.output}"
+        )
+
+    def test_upgrade_bob_layout_change_reconciles_extension_artifacts(self, tmp_path):
+        """Regression (review #3415, 4725829110).
+
+        When a dual-mode agent (Bob) flips layout across an upgrade, the old
+        layout's *extension* artifacts must be reconciled — not left orphaned.
+        A legacy Bob install renders enabled extensions as ``.bob/commands/``
+        command files; migrating to skills via ``--skills`` must remove those
+        command files, recreate the extension as ``.bob/skills/`` skills, and
+        update the extension registry accordingly (and vice-versa for the
+        reverse ``--legacy-commands`` migration).
+        """
+        project = _init_project(
+            tmp_path, "bob", integration_options="--legacy-commands"
+        )
+
+        result = _run_in_project(project, ["extension", "add", "git"])
+        assert result.exit_code == 0, f"extension add failed: {result.output}"
+
+        commands = project / ".bob" / "commands"
+        skills = project / ".bob" / "skills"
+        registry_path = project / ".specify" / "extensions" / ".registry"
+
+        def _git_registry():
+            data = json.loads(registry_path.read_text(encoding="utf-8"))
+            g = data["extensions"]["git"]
+            return list(g.get("registered_commands", {})), g.get(
+                "registered_skills", []
+            )
+
+        # Legacy precondition: git renders as command files under .bob/commands.
+        assert sorted(commands.glob("speckit.git.*.md")), (
+            "legacy Bob should render the git extension as command files"
+        )
+        assert not list(skills.glob("speckit-git-*")) if skills.exists() else True
+        cmds_agents, skill_names = _git_registry()
+        assert "bob" in cmds_agents and not skill_names
+
+        # Migrate legacy -> skills.
+        result = _run_in_project(project, [
+            "integration", "upgrade", "bob",
+            "--integration-options", "--skills",
+            "--script", "sh", "--force",
+        ])
+        assert result.exit_code == 0, f"--skills migration failed: {result.output}"
+
+        # Old-layout git command files removed; skills recreated.
+        assert not sorted(commands.glob("speckit.git.*.md")), (
+            "git extension command files must be removed after --skills migration"
+        )
+        assert sorted(skills.glob("speckit-git-*")), (
+            "git extension must be recreated as skills after --skills migration"
+        )
+        cmds_agents, skill_names = _git_registry()
+        assert "bob" not in cmds_agents, (
+            "extension registry must drop the stale bob command entry"
+        )
+        assert skill_names, "extension registry must record the migrated skills"
+
+        # Migrate skills -> legacy: the reverse reconciliation must also hold.
+        result = _run_in_project(project, [
+            "integration", "upgrade", "bob",
+            "--integration-options", "--legacy-commands",
+            "--script", "sh", "--force",
+        ])
+        assert result.exit_code == 0, (
+            f"--legacy-commands migration failed: {result.output}"
+        )
+        assert not sorted(skills.glob("speckit-git-*")), (
+            "git extension skills must be removed after --legacy-commands migration"
+        )
+        assert sorted(commands.glob("speckit.git.*.md")), (
+            "git extension command files must be recreated in legacy layout"
+        )
+        cmds_agents, skill_names = _git_registry()
+        assert "bob" in cmds_agents and not skill_names
+
+    def test_upgrade_bob_layout_change_rejected_with_presets_installed(self, tmp_path):
+        """Regression (review #3415, 4726193915).
+
+        A command↔skills layout change cannot reconcile preset artifacts (no
+        agent-scoped preset re-registration exists). Rather than silently
+        orphaning preset files / leaving the registry inconsistent, a
+        layout-changing ``upgrade`` must reject the migration with an
+        actionable error *before any mutation* when preset overrides are
+        installed for the agent. A same-layout upgrade must still succeed.
+        """
+        project = _init_project(
+            tmp_path, "bob", integration_options="--legacy-commands"
+        )
+        commands = project / ".bob" / "commands"
+        skills = project / ".bob" / "skills"
+        assert sorted(commands.glob("speckit.*.md"))
+
+        # Simulate an installed preset that registered command overrides for bob.
+        presets_dir = project / ".specify" / "presets"
+        presets_dir.mkdir(parents=True, exist_ok=True)
+        (presets_dir / ".registry").write_text(
+            json.dumps({
+                "presets": {
+                    "my-preset": {
+                        "version": "1.0.0",
+                        "enabled": True,
+                        "registered_commands": {"bob": ["speckit.plan"]},
+                        "registered_skills": [],
+                    }
+                }
+            }),
+            encoding="utf-8",
+        )
+
+        # Layout-changing upgrade is rejected, and nothing is mutated.
+        result = _run_in_project(project, [
+            "integration", "upgrade", "bob",
+            "--integration-options", "--skills",
+            "--script", "sh", "--force",
+        ])
+        assert result.exit_code != 0, "layout change with presets must be rejected"
+        assert "preset" in result.output.lower()
+        assert "my-preset" in result.output
+        assert not skills.exists(), "no skills layout must be scaffolded on rejection"
+        assert sorted(commands.glob("speckit.*.md")), (
+            "legacy command files must be left untouched on rejection"
+        )
+
+        # A same-layout upgrade (no flag) must still succeed with presets present.
+        result = _run_in_project(project, [
+            "integration", "upgrade", "bob", "--script", "sh", "--force",
+        ])
+        assert result.exit_code == 0, (
+            f"same-layout upgrade must not be blocked by presets: {result.output}"
+        )
+
+    def test_upgrade_bob_layout_change_rejected_when_preset_registry_unreadable(
+        self, tmp_path
+    ):
+        """Regression (review #3415, 4744636079).
+
+        The preset guard must fail *closed*: if the preset registry exists but
+        cannot be read/parsed (corruption, permissions), the layout-changing
+        upgrade must be rejected before any mutation rather than proceeding on
+        a false "no presets installed" assumption (which would let ``--force``
+        delete preset-overridden command files while their registry state is
+        unknown). A genuinely absent registry must still be allowed.
+        """
+        project = _init_project(
+            tmp_path, "bob", integration_options="--legacy-commands"
+        )
+        commands = project / ".bob" / "commands"
+        skills = project / ".bob" / "skills"
+        assert sorted(commands.glob("speckit.*.md"))
+
+        # Corrupted (unparseable) registry: exists but cannot be read as JSON.
+        presets_dir = project / ".specify" / "presets"
+        presets_dir.mkdir(parents=True, exist_ok=True)
+        (presets_dir / ".registry").write_text("{ not valid json", encoding="utf-8")
+
+        result = _run_in_project(project, [
+            "integration", "upgrade", "bob",
+            "--integration-options", "--skills",
+            "--script", "sh", "--force",
+        ])
+        assert result.exit_code != 0, (
+            "layout change must be rejected when preset registry is unreadable"
+        )
+        assert "preset registry" in result.output.lower()
+        assert not skills.exists(), "no skills layout may be scaffolded on rejection"
+        assert sorted(commands.glob("speckit.*.md")), (
+            "legacy command files must be untouched when failing closed"
+        )
+
+        # A valid, empty registry must NOT block the migration.
+        (presets_dir / ".registry").write_text(
+            json.dumps({"presets": {}}), encoding="utf-8"
+        )
+        result = _run_in_project(project, [
+            "integration", "upgrade", "bob",
+            "--integration-options", "--skills",
+            "--script", "sh", "--force",
+        ])
+        assert result.exit_code == 0, (
+            f"valid empty preset registry must not block migration: {result.output}"
+        )
+        assert skills.exists(), "skills layout should be scaffolded once unblocked"
+
+    def test_upgrade_secondary_bob_layout_change_preserves_active_agent_skills(
+        self, tmp_path
+    ):
+        """Regression (review #3415, 4726347306).
+
+        ``integration upgrade`` supports upgrading a *secondary* (non-active)
+        integration. The layout-change extension reconciliation must NOT run
+        for a secondary agent: ``unregister_agent_artifacts`` treats the
+        unscoped per-extension ``registered_skills`` as belonging to the passed
+        agent and, if that agent's skills dir is absent, scans every agent's
+        skills dir — which could delete/untrack the *active* agent's extension
+        skills. The following re-registration cannot repair that because
+        extension skill rendering is active-agent-scoped (#2948).
+        """
+        # Active agent: copilot in skills mode → git extension renders as skills.
+        project = _init_project(tmp_path, "copilot", integration_options="--skills")
+        result = _run_in_project(project, ["extension", "add", "git"])
+        assert result.exit_code == 0, f"extension add failed: {result.output}"
+
+        skill = project / ".github" / "skills" / "speckit-git-feature" / "SKILL.md"
+        assert skill.exists(), "precondition: active copilot has the git extension skill"
+
+        registry_path = project / ".specify" / "extensions" / ".registry"
+
+        def _git_skills():
+            data = json.loads(registry_path.read_text(encoding="utf-8"))
+            return data["extensions"]["git"].get("registered_skills", [])
+
+        assert _git_skills(), "precondition: git skills registered for active copilot"
+
+        # Add a secondary (non-active) Bob in the legacy commands layout.
+        result = _run_in_project(project, [
+            "integration", "install", "bob",
+            "--integration-options", "--legacy-commands",
+            "--script", "sh", "--force",
+        ])
+        assert result.exit_code == 0, result.output
+
+        # Flip the *secondary* Bob's layout to skills. copilot stays active.
+        result = _run_in_project(project, [
+            "integration", "upgrade", "bob",
+            "--integration-options", "--skills",
+            "--script", "sh", "--force",
+        ])
+        assert result.exit_code == 0, result.output
+
+        # The active agent's extension skill must be untouched on disk and in
+        # the registry — the secondary layout change must not reconcile it.
+        assert skill.exists(), (
+            "secondary Bob layout change must not delete the active agent's "
+            "extension skill"
+        )
+        assert _git_skills(), (
+            "secondary Bob layout change must not untrack the active agent's "
+            "extension skills in the registry"
+        )
+
     def test_upgrade_preserves_existing_vscode_settings(self, tmp_path):
         """Regression: copilot upgrade must not stale-delete .vscode/settings.json.
 
@@ -2616,6 +2926,82 @@ class TestIntegrationUpgrade:
             "deleted extension skill (#2886)"
         )
 
+    def test_installed_presets_affecting_agent_absent_vs_unreadable(self, tmp_path):
+        """Unit (review #3415, 4744636079): fail closed only when unreadable.
+
+        The preset guard helper must return an empty list for a genuinely
+        absent registry, but raise ``_PresetRegistryUnreadableError`` when the
+        registry exists yet cannot be read/parsed — so a layout-changing
+        upgrade never proceeds on a false "no presets" result.
+        """
+        from specify_cli.integrations._migrate_commands import (
+            _PresetRegistryUnreadableError,
+            _installed_presets_affecting_agent,
+        )
+
+        project = tmp_path / "proj"
+        project.mkdir()
+
+        # Genuinely absent registry → empty list (safe to proceed).
+        assert _installed_presets_affecting_agent(project, "bob") == []
+
+        presets_dir = project / ".specify" / "presets"
+        presets_dir.mkdir(parents=True)
+        registry = presets_dir / ".registry"
+
+        # Corrupted JSON → unreadable → raise.
+        registry.write_text("{ not json", encoding="utf-8")
+        with pytest.raises(_PresetRegistryUnreadableError):
+            _installed_presets_affecting_agent(project, "bob")
+
+        # Malformed structure (presets not a dict) → unreadable → raise.
+        registry.write_text(json.dumps({"presets": []}), encoding="utf-8")
+        with pytest.raises(_PresetRegistryUnreadableError):
+            _installed_presets_affecting_agent(project, "bob")
+
+        # Malformed per-preset entry (not a dict) → ownership unknown → raise.
+        registry.write_text(
+            json.dumps({"presets": {"p1": []}}), encoding="utf-8"
+        )
+        with pytest.raises(_PresetRegistryUnreadableError):
+            _installed_presets_affecting_agent(project, "bob")
+
+        # Malformed registered_commands (not a dict) → raise.
+        registry.write_text(
+            json.dumps({"presets": {"p1": {"registered_commands": []}}}),
+            encoding="utf-8",
+        )
+        with pytest.raises(_PresetRegistryUnreadableError):
+            _installed_presets_affecting_agent(project, "bob")
+
+        # Malformed registered_skills (not a list) → raise.
+        registry.write_text(
+            json.dumps({"presets": {"p1": {"registered_skills": {}}}}),
+            encoding="utf-8",
+        )
+        with pytest.raises(_PresetRegistryUnreadableError):
+            _installed_presets_affecting_agent(project, "bob")
+
+        # Valid, empty registry → empty list.
+        registry.write_text(json.dumps({"presets": {}}), encoding="utf-8")
+        assert _installed_presets_affecting_agent(project, "bob") == []
+
+        # Valid registry with a preset registered for bob → reported.
+        registry.write_text(
+            json.dumps({
+                "presets": {
+                    "p1": {"registered_commands": {"bob": ["speckit.plan"]}},
+                    "p2": {"registered_commands": {"codex": ["speckit.plan"]}},
+                    "p3": {"registered_skills": ["speckit-x"]},
+                }
+            }),
+            encoding="utf-8",
+        )
+        assert sorted(_installed_presets_affecting_agent(project, "bob")) == [
+            "p1",
+            "p3",
+        ]
+
 
 # ── Full lifecycle ───────────────────────────────────────────────────
 
@@ -2733,6 +3119,30 @@ class TestParseIntegrationOptionsEqualsForm:
         assert excinfo.value.exit_code == 1
         assert "Error: Could not parse integration options: No closing quotation." in capsys.readouterr().out
 
+    def test_bad_option_token_with_rich_markup_exits_cleanly(self):
+        """A bad option token carrying Rich markup must exit cleanly, not crash.
+
+        The token is user-controlled and gets interpolated into console.print.
+        A value like '[/red]foo' parses fine through shlex but is an unexpected
+        value / unknown option — and an unbalanced Rich tag would raise
+        rich.errors.MarkupError inside console.print, leaking a traceback
+        instead of the intended typer.Exit(1). The token must be escaped."""
+        import typer
+
+        from specify_cli.integrations._commands import _parse_integration_options
+        from specify_cli.integrations import get_integration
+
+        integration = get_integration("generic")
+        assert integration is not None
+
+        # Unexpected value token carrying markup.
+        with pytest.raises(typer.Exit):
+            _parse_integration_options(integration, "[/red]foo")
+
+        # Unknown option token carrying markup.
+        with pytest.raises(typer.Exit):
+            _parse_integration_options(integration, "--[/red]bad")
+
 
 class TestUninstallNoManifestClearsInitOptions:
     def test_init_options_cleared_on_no_manifest_uninstall(self, tmp_path):

tests/integrations/test_manifest.py

@@ -220,6 +220,28 @@ class TestManifestUninstall:
         m.uninstall()
         assert not m.manifest_path.exists()
 
+    def test_remove_manifest_false_preserves_manifest_file(self, tmp_path):
+        """Regression (review #3415, 4724160183): a partial cleanup must not
+        delete ``{key}.manifest.json``.
+
+        The upgrade stale-file pass builds a throwaway manifest sharing the
+        integration's key over a subset of files and uninstalls it.  With
+        ``remove_manifest=False`` the tracked files are still removed but the
+        real, freshly-saved manifest for that key survives — otherwise a
+        layout-shrinking upgrade (e.g. Bob migrating legacy commands → skills)
+        would leave the integration untracked and un-upgradeable.
+        """
+        m = IntegrationManifest("test", tmp_path, version="1.0")
+        m.record_file("f.txt", "content")
+        m.save()
+        assert m.manifest_path.exists()
+        removed, skipped = m.uninstall(remove_manifest=False)
+        assert len(removed) == 1
+        assert not (tmp_path / "f.txt").exists()
+        assert m.manifest_path.exists(), (
+            "remove_manifest=False must keep the manifest file on disk"
+        )
+
     def test_cleans_empty_parent_dirs(self, tmp_path):
         m = IntegrationManifest("test", tmp_path)
         m.record_file("a/b/c/f.txt", "content")