chore: bump version to 0.14.0

* docs: add spec-kit-copilot to community friends

Assisted-by: GitHub Copilot (model: Claude Opus 4.8, autonomous)

Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Copilot-Session: c4ad8a5e-3ddb-44b2-b8ec-4f438c636b61

* docs: note first-party exception for spec-kit-copilot

Clarify the page disclaimer so it covers first-party GitHub projects, and
mark spec-kit-copilot as a first-party GitHub project.

Assisted-by: GitHub Copilot (model: Claude Opus 4.8, autonomous)

Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Copilot-Session: c4ad8a5e-3ddb-44b2-b8ec-4f438c636b61

---------

Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Copilot-Session: c4ad8a5e-3ddb-44b2-b8ec-4f438c636b61

.devcontainer/devcontainer.json

@@ -65,7 +65,8 @@
 		},
 		"chat.tools.terminal.autoApprove": {
 			".specify/scripts/bash/": true,
-			".specify/scripts/powershell/": true
+			".specify/scripts/powershell/": true,
+			".specify/scripts/python/": true
 		}
       }
     }

.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

AGENTS.md

@@ -187,7 +187,7 @@ context_markers:
   end: "<!-- SPECKIT END -->"
 ```
 
-- The Specify CLI does **not** write this config. When `context_file` is empty, the extension's bundled scripts self-seed it by looking up the active integration's key in the extension's own `agent-context-defaults.json` map (`extensions/agent-context/scripts/bash/update-agent-context.sh` and `.ps1`). The CLI registry is never consulted — all agent→context-file knowledge lives inside the extension.
+- The Specify CLI does **not** write this config. When `context_file` is empty, the extension's bundled scripts self-seed it by looking up the active integration's key in the extension's own `agent-context-defaults.json` map (`extensions/agent-context/scripts/bash/update-agent-context.sh`, `.ps1`, and `extensions/agent-context/scripts/python/update_agent_context.py`). The CLI registry is never consulted — all agent→context-file knowledge lives inside the extension.
 - `context_markers.{start,end}` are read solely by the extension's scripts; they default to the Spec Kit markers shown above and can be customized by editing `agent-context-config.yml` directly.
 
 Existing projects created by older Spec Kit versions keep working: any previously written managed section or extension config is left intact and is only ever updated by the extension when run.
@@ -268,6 +268,25 @@ echo "✅ Done"
 
 ## Command File Formats
 
+### Script References (`scripts:` frontmatter)
+
+Core command templates (`templates/commands/*.md`) that invoke a helper script declare it in a `scripts:` frontmatter block with one line per supported script type. The `{SCRIPT}` placeholder in the command body is replaced at install time with the entry matching the project's selected script type (`--script sh|ps|py`):
+
+```yaml
+scripts:
+  sh: scripts/bash/setup-plan.sh --json
+  ps: scripts/powershell/setup-plan.ps1 -Json
+  py: scripts/python/setup_plan.py --json
+```
+
+| Key  | Script type            | Location                   |
+| ---- | ---------------------- | -------------------------- |
+| `sh` | POSIX shell (bash/zsh) | `scripts/bash/*.sh`        |
+| `ps` | PowerShell             | `scripts/powershell/*.ps1` |
+| `py` | Python                 | `scripts/python/*.py`      |
+
+All three entries must be present and behaviorally equivalent — agents parse the same stdout contract (`FEATURE_DIR:…`, `AVAILABLE_DOCS:…`, `--json` shapes) regardless of which one runs. (The bundled `agent-context` and `git` extension command templates also invoke helpers but do not yet use `scripts:` frontmatter — see [Script Types and Migration](#script-types-and-migration).)
+
 ### Markdown Format
 
 **Standard format:**
@@ -328,9 +347,29 @@ Different agents use different argument placeholders. The placeholder used in co
 - **TOML-based**: `{{args}}` (e.g., Gemini)
 - **YAML-based**: `{{args}}` (e.g., Goose)
 - **Custom**: some agents override the default (e.g., Forge uses `{{parameters}}`)
-- **Script placeholders**: `{SCRIPT}` (replaced with actual script path)
+- **Script placeholders**: `{SCRIPT}` (replaced with the resolved command from the template's `scripts:` frontmatter, per the project's `--script sh|ps|py` selection)
 - **Agent placeholders**: `__AGENT__` (replaced with agent name)
 
+## Script Types and Migration
+
+Spec Kit ships every core workflow script in three interchangeable variants — POSIX shell (`sh`), PowerShell (`ps`), and Python (`py`) — selected per project with `specify init --script sh|ps|py`. Each core command template that invokes a helper script carries all three in its `scripts:` frontmatter (templates that don't call a script, e.g. `constitution`/`specify`, have no `scripts:` block); see [Script References](#script-references-scripts-frontmatter).
+
+### Why Python is recommended
+
+- **No extra runtime.** The `specify` CLI is already Python, so the interpreter is guaranteed present — `py` adds no new dependency.
+- **Path toward a single source of truth.** The shell variants require paired `.sh` + `.ps1` maintenance and diverge on JSON handling (`jq` vs manual parsing). The Python variant avoids `jq` and is intended to eventually replace that dual-maintenance — but that consolidation has not happened yet: all three variants are still maintained in parallel (see the parity rule below).
+- **Parity-tested.** The Python ports are covered by tests — output-parity tests against the shell scripts where the contract is stdout-based, and direct unit tests elsewhere — so the stdout contract agents rely on stays stable.
+
+### Defaults and availability
+
+- `py` is available today for the core command templates (via their `scripts:` frontmatter). The bundled extensions (`agent-context`, `git`) ship Python script variants on disk, but their command templates still hard-code the Bash/PowerShell invocations, so `--script py` does not yet route those extension commands to Python — wiring `py` into the extension command templates is tracked separately.
+- Selection is per project: interactive `specify init` prompts for the script type, while non-interactive runs default to a shell variant by OS (`sh` on Linux/macOS, `ps` on Windows). `py` is chosen at the prompt or via `--script py`.
+- `sh` and `ps` remain fully supported. Nothing is removed, and `py` is not yet the default.
+
+### Parity rule for contributors
+
+All three script types are first-class: any change to a workflow script must update `sh`, `ps`, and `py` together and keep their tests (parity and unit) green. Making `py` the default and eventually retiring `sh`/`ps` is future work gated on adoption, tracked under the script-unification epic ([#3277](https://github.com/github/spec-kit/issues/3277)) — not something to act on from this doc.
+
 ## Special Processing Requirements
 
 Some agents require custom processing beyond the standard template transformations:

CHANGELOG.md

@@ -2,6 +2,56 @@
 
 <!-- insert new changelog below this comment -->
 
+## [0.14.0] - 2026-07-23
+
+### Changed
+
+- docs: add spec-kit-copilot to community friends (#3675)
+- fix(integrations): recompute invoke_separator from retained parsed_options (#3664)
+- fix(workflows): preserve intra-overlay order for multiple insert_after edits (#3662)
+- fix(bundler): reject falsy non-mapping requires/provides in manifest from_dict (#3661)
+- fix(bundler): dump_yaml writes literal UTF-8 (allow_unicode=True) (#3660)
+- fix(integrations): declare kiro-cli multi-install safe (#3477)
+- fix(git-extension): trim trailing whitespace before stripping commit-message quotes (#3673)
+- fix(bundler): order bundle members by canonical POSIX arcname (reproducible builds) (#3658)
+- fix(integrations): Cline overrides post_process_command_content (correct hook name) (#3657)
+- docs(workflows): gate step docstring lists the 'retry' on_reject behaviour (#3656)
+- fix: harden bounded reads and redirect validation (#3671)
+- fix(packaging): bundle scripts/python into the wheel core_pack (#3665) (#3670)
+- fix: bundle scripts/python in wheel so --script py works (#3665) (#3668)
+- docs(workflows): init step docstring lists the 'py' script type (#3655)
+- fix(integrations): declare LingmaIntegration multi_install_safe (#3654)
+- fix: guard constitution command against feature execution (#3646)
+- Fix duplicate step numbering in specify command (#3647)
+- docs(scripts): document the 'py' script type and sh/ps migration plan (#3284) (#3653)
+- harden: bound HTTP reads and enforce strict redirects (#3140)
+- chore: release 0.13.4, begin 0.13.5.dev0 development (#3649)
+
+## [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

docs/community/friends.md

@@ -1,7 +1,7 @@
 # Community Friends
 
 > [!NOTE]
-> Community projects listed here are independently created and maintained by their respective authors. They are **not reviewed, nor endorsed, nor supported by GitHub**. Review their source code before installation and use at your own discretion.
+> Community projects listed here are independently created and maintained by their respective authors. Unless explicitly marked as a **first-party GitHub project**, they are **not reviewed, nor endorsed, nor supported by GitHub**. Review their source code before installation and use at your own discretion.
 
 Community projects that extend, visualize, or build on Spec Kit:
 
@@ -16,3 +16,5 @@ Community projects that extend, visualize, or build on Spec Kit:
 - **[cc-spec-kit](https://github.com/speckit-community/cc-spec-kit)** — Community-maintained plugin for Claude Code and GitHub Copilot CLI that installs Spec Kit skills via the plugin marketplace.
 
 - **[spectatui](https://github.com/tinesoft/spectatui)** — A terminal UI (TUI) dashboard for Spec Kit that lets you track features, manage specifications, integrations, presets, workflows, and extensions, and monitor AI agent workflows. Attach to existing AI sessions or launch new ones from your terminal. Keyboard and mouse support. Light/dark theme support. Customizable and performance-oriented. Requires the `specify` CLI in your PATH.
+
+- **[spec-kit-copilot](https://github.com/github/spec-kit-copilot)** — _First-party GitHub project._ A GitHub Copilot **skills plugin** that exposes the Spec Kit `specify` CLI to the Copilot agent in both the Copilot CLI and the GitHub Copilot app. It provides a focused skill per `specify` command group — setup, init, check, extensions, presets, bundles, workflows, workflow steps, and self-upgrade — so you can navigate and drive the entire Spec Kit ecosystem through natural language, letting Copilot decide when and how to run the right `specify` commands on your behalf.

docs/community/presets.md

@@ -19,6 +19,7 @@ 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) |

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

@@ -2,7 +2,7 @@
 
 This guide shows how to iterate on the `specify` CLI locally without publishing a release or committing to `main` first.
 
-> Scripts now have both Bash (`.sh`) and PowerShell (`.ps1`) variants. The CLI auto-selects based on OS unless you pass `--script sh|ps`.
+> Scripts are available as Bash (`.sh`), PowerShell (`.ps1`), and Python (`.py`) variants. Interactive `specify init` prompts you to choose one; non-interactive runs default to a shell variant for your OS. Pass `--script sh|ps|py` to select explicitly.
 
 ## 1. Clone and Switch Branches
 
@@ -189,7 +189,7 @@ rm -rf .venv dist build *.egg-info
 | `ModuleNotFoundError: typer` | Run `uv pip install -e .` |
 | Scripts not executable (Linux) | Re-run init or `chmod +x scripts/*.sh` |
 | Git commands unavailable | Install the git extension with `specify extension add git` |
-| Wrong script type downloaded | Pass `--script sh` or `--script ps` explicitly |
+| Wrong script type downloaded | Pass `--script sh`, `--script ps`, or `--script py` explicitly |
 | TLS errors on corporate network | Configure your environment's certificate store or proxy. The `--skip-tls` flag is deprecated and has no effect. |
 
 ## 14. Next Steps

docs/quickstart.md

@@ -3,7 +3,7 @@
 This guide will help you get started with Spec-Driven Development using Spec Kit. Throughout, we illustrate each step with a running example: **Taskify**, a small team productivity platform.
 
 > [!NOTE]
-> Automation scripts are provided as both Bash (`.sh`) and PowerShell (`.ps1`) variants. The `specify` CLI auto-selects based on your OS unless you pass `--script sh|ps`.
+> Automation scripts are provided as Bash (`.sh`), PowerShell (`.ps1`), and Python (`.py`) variants. Interactive `specify init` prompts you to choose one; non-interactive runs default to a shell variant for your OS. Pass `--script sh|ps|py` to select explicitly.
 
 > [!NOTE]
 > Commands are shown here in `/speckit.*` form, but the exact invocation depends on your agent. Some skills-based agents use `$speckit-*` (e.g. Codex, ZCode) or `/skill:speckit-*` (e.g. Kimi). Use whichever form your agent exposes — the steps are otherwise identical.

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/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`         |                                                                                                                                           |
@@ -85,7 +86,7 @@ specify integration install <key>
 
 | Option                   | Description                                                              |
 | ------------------------ | ------------------------------------------------------------------------ |
-| `--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)        |
 | `--force`                | Opt in to installing alongside integrations that are not declared multi-install safe |
 | `--integration-options`  | Integration-specific options (e.g. `--integration-options="--commands-dir .myagent/cmds"`) |
 
@@ -121,7 +122,7 @@ specify integration switch <key>
 
 | Option                   | Description                                                              |
 | ------------------------ | ------------------------------------------------------------------------ |
-| `--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)        |
 | `--force`                | Force removal of modified files during uninstall; when the target is already installed, overwrite managed shared templates while changing the default |
 | `--refresh-shared-infra` | Also overwrite shared infrastructure files even if you customized them (otherwise customizations are preserved) |
 | `--integration-options`  | Options for the target integration when it is not already installed      |
@@ -149,7 +150,7 @@ specify integration upgrade [<key>]
 | Option                   | Description                                                              |
 | ------------------------ | ------------------------------------------------------------------------ |
 | `--force`                | Overwrite files even if they have been modified                          |
-| `--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)        |
 | `--integration-options`  | Options for the integration                                              |
 
 Reinstalls an installed integration with updated templates and commands (e.g., after upgrading Spec Kit). Defaults to the default integration; if a key is provided, it must be one of the installed integrations. Detects locally modified files and blocks the upgrade unless `--force` is used. Stale files from the previous install that are no longer needed are removed automatically. Shared templates stay aligned with the default integration even when upgrading a non-default integration.
@@ -256,31 +257,30 @@ Spec Kit tracks one default integration in `.specify/integration.json` with `def
 
 An integration is multi-install safe when it uses a static, unique agent root and command directory, stable command invocation settings, and a separate install manifest whose managed files do not overlap another safe integration. Registry tests enforce those path and manifest invariants. Shared Spec Kit templates remain aligned to the single default integration.
 
-The Isolation column below lists paths Spec Kit manages for that integration (skills/commands roots and any integration-owned rule files). It is not a full inventory of every file an agent may read.
-
-**Agent-context defaults are separate.** The optional agent-context extension maps each integration to a default context file in `extensions/agent-context/agent-context-defaults.json`. Those defaults are independent of multi-install safety: several agents may share a root file such as `AGENTS.md` when the extension is enabled. Multi-install safety does not require a unique context file per safe integration.
+The Command directory column below lists the directory each integration installs its commands or skills into. Context-file targeting is a separate concern from integration multi-install safety: `multi_install_safe` is an integration declaration about command/skill paths, whereas the optional agent-context extension manages a per-agent context file (for example `AGENTS.md` or `CLAUDE.md`) and can even synchronize several anchors at once via its `context_files` setting. Multiple agents mapping to the same context file is expected there and does not affect whether an integration is multi-install safe; see the agent-context extension for details.
 
 The currently declared multi-install safe integrations are:
 
-| Key | Isolation |
-| --- | --------- |
-| `auggie` | `.augment/commands`, `.augment/rules/specify-rules.md` |
-| `claude` | `.claude/skills`, `CLAUDE.md` |
-| `cline` | `.clinerules/workflows`, `.clinerules/specify-rules.md` |
-| `codebuddy` | `.codebuddy/commands`, `CODEBUDDY.md` |
-| `codex` | `.agents/skills`, `AGENTS.md` |
-| `cursor-agent` | `.cursor/skills`, `.cursor/rules/specify-rules.mdc` |
-| `firebender` | `.firebender/commands`, `.firebender/rules/specify-rules.mdc` |
-| `gemini` | `.gemini/commands`, `GEMINI.md` |
+| Key | Command directory |
+| --- | ----------------- |
+| `auggie` | `.augment/commands` |
+| `claude` | `.claude/skills` |
+| `cline` | `.clinerules/workflows` |
+| `codebuddy` | `.codebuddy/commands` |
+| `codex` | `.agents/skills` |
+| `cursor-agent` | `.cursor/skills` |
+| `firebender` | `.firebender/commands` |
+| `gemini` | `.gemini/commands` |
 | `grok` | `.grok/skills` |
-| `junie` | `.junie/commands`, `.junie/AGENTS.md` |
-| `kilocode` | `.kilocode/workflows`, `.kilocode/rules/specify-rules.md` |
-| `qodercli` | `.qoder/commands`, `QODER.md` |
-| `qwen` | `.qwen/commands`, `QWEN.md` |
-| `shai` | `.shai/commands`, `SHAI.md` |
-| `tabnine` | `.tabnine/agent/commands`, `TABNINE.md` |
-| `trae` | `.trae/skills`, `.trae/rules/project_rules.md` |
-| `zcode` | `.zcode/skills`, `ZCODE.md` |
+| `junie` | `.junie/commands` |
+| `kilocode` | `.kilocode/workflows` |
+| `kiro-cli` | `.kiro/prompts` |
+| `qodercli` | `.qoder/commands` |
+| `qwen` | `.qwen/commands` |
+| `shai` | `.shai/commands` |
+| `tabnine` | `.tabnine/agent/commands` |
+| `trae` | `.trae/skills` |
+| `zcode` | `.zcode/skills` |
 
 Integrations that share a command directory with another integration, require dynamic install paths such as `--commands-dir`, or merge shared tool settings are not declared safe by default. They can still be installed alongside another integration with `--force`.
 

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/git/scripts/bash/auto-commit.sh

@@ -94,7 +94,12 @@ if [ -f "$_config_file" ]; then
                     [ "$_val" = "false" ] && _enabled=false
                 fi
                 if echo "$_line" | grep -Eq '[[:space:]]+message:'; then
-                    _commit_msg=$(echo "$_line" | sed 's/^[^:]*:[[:space:]]*//' | sed 's/^["'\'']//' | sed 's/["'\'']*$//')
+                    # Trim trailing whitespace before stripping the closing quote:
+                    # a value like `message: "Done"  ` (trailing spaces after the
+                    # quote) would otherwise leave the quote dangling (`Done"  `),
+                    # since the closing-quote strip is anchored to end-of-string.
+                    # The PowerShell twin .Trim()s first; match it for parity.
+                    _commit_msg=$(echo "$_line" | sed 's/^[^:]*:[[:space:]]*//' | sed 's/[[:space:]]*$//' | sed 's/^["'\'']//' | sed 's/["'\'']*$//')
                 fi
             fi
         fi

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"
     }
 }

extensions/git/scripts/python/auto_commit.py

@@ -33,7 +33,15 @@ def _value_after_colon(line: str) -> str:
 
 
 def _strip_quotes(value: str) -> str:
-    """Strip one leading quote and all trailing quotes, mirroring the bash sed."""
+    """Strip surrounding whitespace, then one leading quote and all trailing quotes.
+
+    Trimming first matters when the YAML value has trailing whitespace after a
+    closing quote (``message: "Done"  ``): stripping quotes anchored to the end
+    of string would leave the closing quote dangling (``Done"  ``) because the
+    quote is no longer at the end. The PowerShell twin ``.Trim()``s before
+    stripping, so trim here too to keep all three script variants in parity.
+    """
+    value = value.strip()
     value = re.sub(r"^[\"']", "", value)
     return re.sub(r"[\"']*$", "", value)
 

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",

presets/catalog.community.json

@@ -364,6 +364,35 @@
       "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",

presets/lean/commands/speckit.constitution.md

@@ -8,6 +8,24 @@ description: Create or update the project constitution.
 $ARGUMENTS
 ```
 
+## Scope Guard
+
+This command's own work is limited to creating or updating the project constitution and
+propagating constitution-driven changes to dependent Spec Kit artifacts.
+
+- Classify every part of the user input as constitution content or a separate non-governance
+  intent. Feature implementation, code generation, refactoring, build, and deployment requests
+  are examples of non-governance intents.
+- You **MUST NOT** execute any non-governance intent. Defer each one to `Next Actions`.
+- You **MUST NOT** create, modify, or delete application source files or other artifacts
+  unrelated to the constitution workflow.
+- If an instruction could be either constitution content or a non-governance intent, ask for
+  clarification before making changes.
+- After updating the constitution, list each deferred intent in a `Next Actions` section with an
+  appropriate follow-up Spec Kit command, such as `__SPECKIT_COMMAND_SPECIFY__`, but do not
+  invoke it.
+- Omit `Next Actions` when there are no non-governance intents.
+
 ## Outline
 
 1. Create or update the project constitution and store it in `.specify/memory/constitution.md`.

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "specify-cli"
-version = "0.13.3"
+version = "0.14.0"
 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"
@@ -39,6 +39,7 @@ packages = ["src/specify_cli"]
 "templates/commands" = "specify_cli/core_pack/commands"
 "scripts/bash" = "specify_cli/core_pack/scripts/bash"
 "scripts/powershell" = "specify_cli/core_pack/scripts/powershell"
+"scripts/python" = "specify_cli/core_pack/scripts/python"
 # Bundled extensions (installable via `specify extension add <name>`)
 "extensions/git" = "specify_cli/core_pack/extensions/git"
 "extensions/agent-context" = "specify_cli/core_pack/extensions/agent-context"

src/specify_cli/_download_security.py

@@ -0,0 +1,218 @@
+"""Helpers for bounded HTTP downloads."""
+
+from __future__ import annotations
+
+import io
+import socket
+from ipaddress import IPv4Address, IPv6Address, ip_address
+from typing import NoReturn, TypeVar
+from urllib.parse import ParseResult, urlparse
+
+
+ErrorT = TypeVar("ErrorT", bound=Exception)
+
+MAX_DOWNLOAD_BYTES = 50 * 1024 * 1024
+READ_CHUNK_SIZE = 64 * 1024
+
+# Tighter ceiling for responses that are read fully into memory and parsed as
+# JSON. The 50 MiB MAX_DOWNLOAD_BYTES default is sized for archive/payload
+# downloads; JSON metadata responses are far smaller, so capping them close to
+# their real size shrinks the memory-DoS surface and keeps the "too large"
+# error reachable (rather than only triggering on tens of MiB). Pass it
+# explicitly at each JSON call site so the intended bound is pinned there.
+# METADATA covers fixed-shape single-object responses (an OAuth token, one
+# release's metadata): a few KiB in practice, 1 MiB is already generous.
+MAX_JSON_METADATA_BYTES = 1 * 1024 * 1024
+
+
+def _ip_address_without_scope(
+    hostname: str,
+) -> IPv4Address | IPv6Address | None:
+    """Parse a canonical IP literal, validating an optional IPv6 zone ID."""
+    if "%" in hostname:
+        # Accept only the RFC 6874 ``%25<zone>`` spelling. Other escapes can
+        # alter the IPv6 address when urllib unquotes the authority.
+        address_text, separator, zone = hostname.partition("%25")
+        if (
+            not separator
+            or ":" not in address_text
+            or "%" in address_text
+            or "%" in zone
+        ):
+            return None
+        if not zone or any(
+            not (character.isascii() and (character.isalnum() or character in "._~-"))
+            for character in zone
+        ):
+            return None
+    else:
+        address_text = hostname
+    try:
+        address = ip_address(address_text)
+    except ValueError:
+        return None
+    if "%" in hostname and not isinstance(address, IPv6Address):
+        return None
+    return address
+
+
+def _is_ip_loopback(address: IPv4Address | IPv6Address | None) -> bool:
+    if address is None:
+        return False
+    mapped = getattr(address, "ipv4_mapped", None)
+    return address.is_loopback or bool(mapped and mapped.is_loopback)
+
+
+def _is_ip_local_redirect_target(
+    address: IPv4Address | IPv6Address | None,
+) -> bool:
+    """Treat loopback and unspecified listener aliases as local targets."""
+    if address is None:
+        return False
+    mapped = getattr(address, "ipv4_mapped", None)
+    return _is_ip_loopback(address) or address.is_unspecified or bool(
+        mapped and mapped.is_unspecified
+    )
+
+
+def _parse_url(url: str) -> ParseResult | None:
+    """Parse *url*, rejecting missing hosts and malformed ports."""
+    try:
+        parsed = urlparse(url)
+        hostname = parsed.hostname
+        # Accessing ``port`` performs urllib's range and syntax validation.
+        parsed.port
+    except (TypeError, ValueError):
+        return None
+    if not hostname:
+        return None
+
+    if "%" in hostname:
+        # urllib unquotes reg-name/IPv4 authorities before connecting. Reject
+        # them so encoded dots, characters, ports, or brackets cannot make the
+        # validated hostname differ from the effective target. The only safe
+        # percent form retained is a validated bracketed IPv6 zone ID.
+        if _ip_address_without_scope(hostname) is None:
+            return None
+    elif ":" not in hostname:
+        try:
+            hostname.encode("idna")
+        except UnicodeError:
+            return None
+    return parsed
+
+
+def _is_definite_loopback_host(hostname: str) -> bool:
+    """Recognize only unambiguous hosts that may safely authorize HTTP."""
+    if not hostname.isascii():
+        return False
+    if hostname == "localhost":
+        return True
+    return _is_ip_loopback(_ip_address_without_scope(hostname))
+
+
+def _is_potential_local_target_host(hostname: str) -> bool:
+    """Conservatively classify aliases that could reach a local listener."""
+    if ":" in hostname:
+        return _is_ip_local_redirect_target(_ip_address_without_scope(hostname))
+    try:
+        host = hostname.encode("idna").decode("ascii").lower().removesuffix(".")
+    except UnicodeError:
+        return False
+    if host == "localhost" or host.endswith(".localhost"):
+        return True
+
+    address = _ip_address_without_scope(host)
+    if address is None:
+        # Historical IPv4 spellings are resolver-dependent. They are never
+        # trusted to authorize HTTP, but treating them as potentially local
+        # prevents them from bypassing a remote-to-loopback redirect check.
+        try:
+            address = ip_address(socket.inet_aton(host))
+        except OSError:
+            return False
+    return _is_ip_local_redirect_target(address)
+
+
+def is_loopback_url(url: str) -> bool:
+    """Return whether *url* has an unambiguous loopback host."""
+    parsed = _parse_url(url)
+    return parsed is not None and _is_definite_loopback_host(parsed.hostname)
+
+
+def _is_potential_local_target_url(url: str) -> bool:
+    parsed = _parse_url(url)
+    return parsed is not None and _is_potential_local_target_host(parsed.hostname)
+
+
+def is_https_or_localhost_http(url: str) -> bool:
+    """Return True if *url* is HTTPS, or HTTP limited to loopback hosts.
+
+    Shared scheme-safety predicate used by the auth HTTP redirect handler and
+    direct URL validations in CLI download flows.
+
+    A hostname is always required: a URL without one (e.g. ``https:///x``)
+    has no real target and is rejected regardless of scheme.
+
+    The HTTP exception is deliberately limited to unambiguous ``localhost``
+    and canonical IPv4/IPv6 loopback literals. Ambiguous numeric, Unicode, and
+    unspecified-address aliases are classified defensively for redirects but
+    never authorize HTTP. No DNS lookup is performed; DNS and hosts-file
+    aliases require connection-level rebinding protection outside this helper.
+    """
+    parsed = _parse_url(url)
+    if parsed is None:
+        return False
+    return parsed.scheme == "https" or (
+        parsed.scheme == "http" and _is_definite_loopback_host(parsed.hostname)
+    )
+
+
+def is_safe_download_redirect(old_url: str, new_url: str) -> bool:
+    """Return whether a redirect preserves the shared download URL policy."""
+    if not is_https_or_localhost_http(new_url):
+        return False
+    return not _is_potential_local_target_url(new_url) or is_loopback_url(old_url)
+
+
+def _raise(error_type: type[ErrorT], message: str) -> NoReturn:
+    raise error_type(message)
+
+
+def read_response_limited(
+    response,
+    *,
+    max_bytes: int = MAX_DOWNLOAD_BYTES,
+    error_type: type[ErrorT] = ValueError,
+    label: str = "download",
+) -> bytes:
+    """Read at most *max_bytes* from a response object.
+
+    ``response.read(n)`` is only guaranteed to return *up to* ``n`` bytes and may
+    return fewer even when more data is pending (e.g. chunked transfer encoding),
+    so a single ``read(max_bytes + 1)`` cannot enforce the bound on its own. Read
+    in a loop until EOF or until one byte past the limit has been accumulated.
+
+    *max_bytes* is keyword-only. It defaults to the module-wide
+    ``MAX_DOWNLOAD_BYTES`` (50 MiB) ceiling for archive/payload downloads;
+    callers with a tighter budget (e.g. small JSON responses) should pass an
+    explicit value so the intended bound is pinned at the call site rather than
+    tracking changes to the shared default.
+    """
+    if isinstance(max_bytes, bool) or not isinstance(max_bytes, int):
+        raise TypeError("max_bytes must be an integer")
+    if max_bytes < 0:
+        raise ValueError("max_bytes must be non-negative")
+
+    output = io.BytesIO()
+    total = 0
+    limit = max_bytes + 1
+    while total < limit:
+        chunk = response.read(min(READ_CHUNK_SIZE, limit - total))
+        if not chunk:
+            break
+        total += len(chunk)
+        if total > max_bytes:
+            _raise(error_type, f"{label} exceeds maximum size of {max_bytes} bytes")
+        output.write(chunk)
+    return output.getvalue()

src/specify_cli/_github_http.py

@@ -100,6 +100,8 @@ def resolve_github_release_asset_api_url(
     import json
     import urllib.error
 
+    from specify_cli._download_security import read_response_limited
+
     parsed = urlparse(download_url)
     hostname = (parsed.hostname or "").lower()
     parts = [unquote(part) for part in parsed.path.strip("/").split("/")]
@@ -158,10 +160,13 @@ def resolve_github_release_asset_api_url(
         if redirect_validator is not None:
             open_kwargs["redirect_validator"] = redirect_validator
         with open_url_fn(release_url, **open_kwargs) as response:
-            raw_release_data = response.read(max_metadata_bytes + 1)
-            if len(raw_release_data) > max_metadata_bytes:
-                raise ValueError("GitHub release metadata exceeds size limit")
-            release_data = json.loads(raw_release_data)
+            release_data = json.loads(
+                read_response_limited(
+                    response,
+                    max_bytes=max_metadata_bytes,
+                    label=f"GitHub release metadata {release_url}",
+                )
+            )
     except (
         urllib.error.URLError,
         json.JSONDecodeError,

src/specify_cli/_version.py

@@ -4,8 +4,8 @@ Pure helpers for comparing PEP 440 versions and fetching the latest GitHub
 release tag.  The ``self_app`` Typer sub-command group is co-located here so
 all version-related logic lives in one place.
 
-Dependencies: stdlib + packaging + ._console only (no other internal imports
-at module level, keeping this layer thin and circular-import-safe).
+Dependencies: stdlib + packaging + ._console + ._download_security only
+(keeping this layer thin and circular-import-safe).
 """
 from __future__ import annotations
 
@@ -28,6 +28,7 @@ from pathlib import Path
 import typer
 from packaging.version import InvalidVersion, Version
 
+from ._download_security import MAX_JSON_METADATA_BYTES, read_response_limited
 from ._console import console
 
 GITHUB_API_LATEST = "https://api.github.com/repos/github/spec-kit/releases/latest"
@@ -119,7 +120,13 @@ def _fetch_latest_release_tag() -> tuple[str | None, str | None]:
             timeout=5,
             extra_headers={"Accept": "application/vnd.github+json"},
         ) as resp:
-            payload = json.loads(resp.read().decode("utf-8"))
+            payload = json.loads(
+                read_response_limited(
+                    resp,
+                    max_bytes=MAX_JSON_METADATA_BYTES,
+                    label="GitHub latest release",
+                ).decode("utf-8")
+            )
             tag = payload.get("tag_name")
             if not isinstance(tag, str) or not tag:
                 raise ValueError("GitHub API response missing valid tag_name")

src/specify_cli/authentication/azure_devops.py

@@ -8,6 +8,7 @@ import os
 import subprocess
 from typing import TYPE_CHECKING
 
+from .._download_security import MAX_JSON_METADATA_BYTES, read_response_limited
 from .base import AuthProvider
 
 if TYPE_CHECKING:
@@ -17,6 +18,20 @@ if TYPE_CHECKING:
 _ADO_RESOURCE_ID = "499b84ac-1321-427f-aa17-267ca6975798"
 
 
+class _TokenResponseTooLarge(Exception):
+    """Raised when an Azure AD token response exceeds the bounded read limit."""
+
+
+def _extract_token(payload: object, key: str) -> str | None:
+    """Return a normalized token from a JSON object, or None for other shapes."""
+    if not isinstance(payload, dict):
+        return None
+    token = payload.get(key)
+    if not isinstance(token, str):
+        return None
+    return token.strip() or None
+
+
 class AzureDevOpsAuth(AuthProvider):
     """Azure DevOps authentication provider.
 
@@ -74,8 +89,7 @@ class AzureDevOpsAuth(AuthProvider):
             if result.returncode != 0:
                 return None
             payload = _json.loads(result.stdout)
-            token = payload.get("accessToken", "").strip()
-            return token or None
+            return _extract_token(payload, "accessToken")
         except (
             OSError,
             subprocess.TimeoutExpired,
@@ -119,9 +133,37 @@ class AzureDevOpsAuth(AuthProvider):
             headers={"Content-Type": "application/x-www-form-urlencoded"},
         )
         try:
-            with urllib.request.urlopen(req, timeout=30) as resp:  # noqa: S310
-                payload = _json.loads(resp.read().decode("utf-8"))
-                token = payload.get("access_token", "").strip()
-                return token or None
-        except (urllib.error.URLError, OSError, _json.JSONDecodeError, KeyError):
+            from specify_cli.authentication.http import _StripAuthOnRedirect
+
+            def reject_token_redirect(_old_url: str, new_url: str) -> None:
+                # A 307/308 redirect preserves this POST body, including the
+                # client_secret. Refuse every redirect so credentials cannot
+                # leave the fixed Microsoft token endpoint.
+                raise urllib.error.URLError(
+                    f"Azure AD token request must not be redirected to {new_url}"
+                )
+
+            opener = urllib.request.build_opener(
+                _StripAuthOnRedirect((), reject_token_redirect)
+            )
+            with opener.open(req, timeout=30) as resp:  # noqa: S310
+                payload = _json.loads(
+                    read_response_limited(
+                        resp,
+                        max_bytes=MAX_JSON_METADATA_BYTES,
+                        error_type=_TokenResponseTooLarge,
+                        label="Azure DevOps token response",
+                    ).decode("utf-8")
+                )
+                return _extract_token(payload, "access_token")
+        except (
+            urllib.error.URLError,
+            OSError,
+            _json.JSONDecodeError,
+            UnicodeDecodeError,
+            _TokenResponseTooLarge,
+        ):
+            # Network failure, malformed JSON, or an oversized response — fall
+            # through to the next strategy. Unrelated programming errors (other
+            # ValueErrors, KeyErrors) intentionally propagate so they surface.
             return None

src/specify_cli/authentication/http.py

@@ -17,6 +17,7 @@ from fnmatch import fnmatch
 from typing import Callable
 from urllib.parse import urlparse
 
+from .._download_security import is_safe_download_redirect
 from . import get_provider
 from .config import AuthConfigEntry, _default_config_path, find_entries_for_url, load_auth_config
 
@@ -60,8 +61,23 @@ def _hostname_in_hosts(hostname: str, hosts: tuple[str, ...]) -> bool:
 RedirectValidator = Callable[[str, str], None]
 
 
+def _validate_strict_redirect(old_url: str, new_url: str) -> None:
+    if not is_safe_download_redirect(old_url, new_url):
+        raise urllib.error.URLError(
+            f"unsafe redirect to {new_url}: target must use HTTPS with a hostname, "
+            "must not enter a local target from a remote host, and may use HTTP only "
+            "within loopback (for example localhost, 127.0.0.1, ::1)"
+        )
+
+
 class _StripAuthOnRedirect(urllib.request.HTTPRedirectHandler):
-    """Drop ``Authorization`` when a redirect leaves trusted hosts or downgrades."""
+    """Redirect handler that guards every redirect it is installed for.
+
+    1. Run any caller-provided redirect validator.
+    2. Reject redirects that are not HTTPS with a hostname. HTTP loopback is
+       allowed only when the previous hop is also loopback.
+    3. Drop ``Authorization`` when a redirect leaves trusted hosts or downgrades.
+    """
 
     def __init__(
         self,
@@ -75,6 +91,8 @@ class _StripAuthOnRedirect(urllib.request.HTTPRedirectHandler):
     def redirect_request(self, req, fp, code, msg, headers, newurl):
         try:
             new_parsed = urlparse(newurl)
+            # Force urllib's syntax and range validation before following.
+            new_parsed.port
         except ValueError as exc:
             # Malformed redirect target (e.g. unterminated IPv6 bracket).
             # Surface as URLError so callers' download error handling applies.
@@ -82,6 +100,7 @@ class _StripAuthOnRedirect(urllib.request.HTTPRedirectHandler):
 
         if self._redirect_validator is not None:
             self._redirect_validator(req.full_url, newurl)
+        _validate_strict_redirect(req.full_url, newurl)
 
         original_auth = (
             req.get_header("Authorization")
@@ -155,6 +174,12 @@ def open_url(
     *extra_headers* (e.g. ``Accept``) are merged into every attempt.
     *redirect_validator*, when provided, is called with ``(old_url, new_url)``
     before following each redirect and may raise to reject the redirect.
+
+    Every attempt uses an isolated opener so a process-wide opener installed
+    with ``urllib.request.install_opener`` cannot replace the redirect guard.
+    Redirect scheme safety: every attempt goes through
+    ``_StripAuthOnRedirect``, which rejects redirects to non-HTTPS URLs except
+    HTTP between loopback URLs, and rejects remote-to-local redirects.
     """
     entries = find_entries_for_url(url, _load_config())
 
@@ -188,7 +213,7 @@ def open_url(
 
     # No entry worked (or none matched) — unauthenticated fallback
     req = _make_req({})
-    if redirect_validator is not None:
-        opener = urllib.request.build_opener(_StripAuthOnRedirect((), redirect_validator))
-        return opener.open(req, timeout=timeout)
-    return urllib.request.urlopen(req, timeout=timeout)  # noqa: S310
+    # No auth is attached on this path, so the handler's host list is empty:
+    # here it runs redirect validation only, not auth stripping.
+    opener = urllib.request.build_opener(_StripAuthOnRedirect((), redirect_validator))
+    return opener.open(req, timeout=timeout)

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/lib/yamlio.py

@@ -60,7 +60,13 @@ def dump_yaml(path: Path, data: Any, *, within: Path | None = None) -> Path:
     try:
         path.parent.mkdir(parents=True, exist_ok=True)
         with path.open("w", encoding="utf-8") as handle:
-            yaml.safe_dump(data, handle, sort_keys=False, default_flow_style=False)
+            yaml.safe_dump(
+                data,
+                handle,
+                sort_keys=False,
+                default_flow_style=False,
+                allow_unicode=True,
+            )
     except OSError as exc:
         raise BundlerError(f"Could not write {path}: {exc}") from exc
     return path

src/specify_cli/bundler/models/catalog.py

@@ -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

@@ -111,8 +111,10 @@ class BundleManifest:
             license=str(bundle_raw.get("license", "")).strip(),
         )
 
-        requires_raw = data.get("requires") or {}
-        if not isinstance(requires_raw, dict):
+        requires_raw = data.get("requires")
+        if requires_raw is None:
+            requires_raw = {}
+        elif not isinstance(requires_raw, dict):
             raise BundlerError("'requires' must be a mapping when present.")
         requires = Requires(
             speckit_version=str(requires_raw.get("speckit_version", "")).strip(),
@@ -122,11 +124,18 @@ 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())
 
-        provides = data.get("provides") or {}
-        if not isinstance(provides, dict):
+        provides = data.get("provides")
+        if provides is None:
+            provides = {}
+        elif not isinstance(provides, dict):
             raise BundlerError("'provides' must be a mapping when present.")
 
         tags_raw = data.get("tags")

src/specify_cli/bundler/services/packager.py

@@ -142,4 +142,10 @@ def _collect_files(
                 # Skip symlinked files to avoid escaping the bundle directory.
                 continue
             collected.append(path)
-    return sorted(collected)
+    # Order by the canonical POSIX arcname (the same key build_bundle uses to
+    # NAME each member), not by pathlib.Path comparison. Path ordering is
+    # platform-dependent (Windows folds case and uses backslash separators),
+    # which would lay out zip members differently across build hosts and break
+    # the byte-for-byte reproducible-build guarantee even though the member
+    # names are identical.
+    return sorted(collected, key=lambda p: p.relative_to(bundle_dir).as_posix())

src/specify_cli/commands/init.py

@@ -700,6 +700,7 @@ 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
@@ -776,6 +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

@@ -3608,6 +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:
@@ -3618,6 +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

@@ -64,8 +64,15 @@ def with_integration_setting(
     elif raw_options is not None:
         current.pop("parsed_options", None)
 
+    # Recompute the separator from the options actually STORED on ``current``
+    # after the update, not the raw ``parsed_options`` argument. When only
+    # ``script_type`` changes (``parsed_options`` and ``raw_options`` both
+    # None), the previously-stored ``parsed_options`` are retained above, so
+    # deriving the separator from the argument (None) would drop an
+    # options-dependent separator (e.g. Copilot ``--skills`` -> "-") back to
+    # the default ".".
     current["invoke_separator"] = integration.effective_invoke_separator(
-        parsed_options, project_root
+        current.get("parsed_options"), project_root
     )
     settings[key] = current
     return settings

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/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:

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)
@@ -125,8 +138,14 @@ class ClineIntegration(MarkdownIntegration):
             content,
         )
 
-    def post_process_content(self, content: str) -> str:
-        """Apply Cline-specific transformations to command content."""
+    def post_process_command_content(self, content: str) -> str:
+        """Apply Cline-specific transformations to command content.
+
+        Overrides the ``IntegrationBase`` hook of the same name so that
+        ``CommandRegistrar.register_commands()`` (which dispatches to
+        ``post_process_command_content``) applies these transforms to
+        extension/preset command files too, not just core commands.
+        """
         updated = self._inject_hook_command_note(content)
         updated = self._rewrite_handoff_references(updated)
         return updated
@@ -156,7 +175,7 @@ class ClineIntegration(MarkdownIntegration):
             content_bytes = path.read_bytes()
             content = content_bytes.decode("utf-8")
 
-            updated = self.post_process_content(content)
+            updated = self.post_process_command_content(content)
 
             if updated != content:
                 path.write_bytes(updated.encode("utf-8"))

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/kiro_cli/__init__.py

@@ -13,6 +13,13 @@ _KIRO_ARG_FALLBACK = "(the user will provide the argument in this conversation)"
 
 class KiroCliIntegration(MarkdownIntegration):
     key = "kiro-cli"
+    # Kiro CLI keeps everything under a static, isolated agent root
+    # (``.kiro/`` with commands in ``.kiro/prompts``) that no other
+    # integration writes to, so it is safe to install alongside others
+    # (issue #3471). IntegrationBase defaults this to False; declaring it
+    # True here is the actual behavior change this integration opts into.
+    # The registry's multi-install-safe contract tests enforce that
+    # isolation for every integration setting this flag.
     multi_install_safe = True
     config = {
         "name": "Kiro CLI",
@@ -27,10 +34,3 @@ class KiroCliIntegration(MarkdownIntegration):
         "args": _KIRO_ARG_FALLBACK,
         "extension": ".md",
     }
-
-    # Kiro CLI keeps everything under a static, isolated agent root
-    # (``.kiro/`` with commands in ``.kiro/prompts``) that no other
-    # integration writes to, so it is safe to install alongside others
-    # (issue #3471). The registry's multi-install-safe contract tests
-    # enforce that isolation for every integration setting this flag.
-    multi_install_safe = True

src/specify_cli/integrations/lingma/__init__.py

@@ -27,6 +27,7 @@ class LingmaIntegration(SkillsIntegration):
         "args": "$ARGUMENTS",
         "extension": "/SKILL.md",
     }
+    multi_install_safe = True
 
     @classmethod
     def options(cls) -> list[IntegrationOption]:

src/specify_cli/presets/_commands.py

@@ -16,6 +16,10 @@ import yaml
 from rich.markup import escape as _escape_markup
 
 from .._console import console
+from .._download_security import (
+    is_https_or_localhost_http,
+    is_safe_download_redirect,
+)
 
 preset_app = typer.Typer(
     name="preset",
@@ -102,38 +106,25 @@ def preset_add(
 
         elif from_url:
             # Validate URL scheme before downloading
-            from ipaddress import ip_address
             from urllib.parse import urlparse as _urlparse
 
             try:
                 _parsed = _urlparse(from_url)
+                _parsed.port
             except ValueError:
                 console.print(f"[red]Error:[/red] Invalid URL: {_escape_markup(from_url)}")
                 raise typer.Exit(1)
 
-            def _is_allowed_download_url(parsed_url):
-                host = parsed_url.hostname
-                if not host:
-                    return False
-                is_loopback = host == "localhost"
-                if not is_loopback:
-                    try:
-                        is_loopback = ip_address(host).is_loopback
-                    except ValueError:
-                        # Host is not an IP literal (e.g., a regular hostname); treat as non-loopback.
-                        pass
-                return parsed_url.scheme == "https" or (parsed_url.scheme == "http" and is_loopback)
-
             def _validate_download_redirect(old_url, new_url):
-                if not _is_allowed_download_url(_urlparse(new_url)):
+                if not is_safe_download_redirect(old_url, new_url):
                     import urllib.error
 
                     raise urllib.error.URLError(
-                        "redirect target must use HTTPS with a hostname, "
-                        "or HTTP for localhost/loopback"
+                        "redirect target must use HTTPS without entering a local "
+                        "target, or stay within loopback over HTTP"
                     )
 
-            if not _is_allowed_download_url(_parsed):
+            if not is_https_or_localhost_http(from_url):
                 console.print(
                     "[red]Error:[/red] URL must use HTTPS with a hostname, "
                     "or HTTP for localhost/loopback."
@@ -167,7 +158,7 @@ def preset_add(
                         redirect_validator=_validate_download_redirect,
                     ) as response:
                         final_url = response.geturl() if hasattr(response, "geturl") else from_url
-                        if not _is_allowed_download_url(_urlparse(final_url)):
+                        if not is_https_or_localhost_http(final_url):
                             console.print(
                                 "[red]Error:[/red] Preset URL redirected to a disallowed URL: "
                                 f"{final_url}. Redirect targets must use HTTPS with a hostname, "

src/specify_cli/workflows/_commands.py

@@ -20,6 +20,10 @@ import yaml
 from rich.markup import escape as _escape_markup
 
 from .._console import console, err_console
+from .._download_security import (
+    is_https_or_localhost_http,
+    is_safe_download_redirect,
+)
 from .._project import _resolve_init_dir_override
 
 workflow_app = typer.Typer(
@@ -383,27 +387,12 @@ _RESERVED_WORKFLOW_IDS: frozenset[str] = frozenset({"overlays", "runs", "steps"}
 def _reject_insecure_download_redirect(old_url: str, new_url: str) -> None:
     """Reject insecure redirects before they are followed."""
     import urllib.error
-    from ipaddress import ip_address
-    from urllib.parse import urlparse
 
-    def _is_loopback_http(url: str) -> bool:
-        parsed = urlparse(url)
-        if parsed.scheme != "http":
-            return False
-        host = parsed.hostname or ""
-        if host == "localhost":
-            return True
-        try:
-            return ip_address(host).is_loopback
-        except ValueError:
-            return False
-
-    if urlparse(new_url).scheme == "https":
-        return
-    if _is_loopback_http(old_url) and _is_loopback_http(new_url):
+    if is_safe_download_redirect(old_url, new_url):
         return
     raise urllib.error.URLError(
-        "redirect target must use HTTPS; loopback HTTP may only redirect from loopback HTTP"
+        "redirect target must use HTTPS without entering a local target; "
+        "loopback HTTP may only redirect from another loopback URL"
     )
 
 
@@ -1555,7 +1544,7 @@ 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():
@@ -1579,24 +1568,15 @@ def workflow_add(
         else (source if source.startswith(("http://", "https://")) else None)
     )
     if download_url is not None:
-        from ipaddress import ip_address
         from urllib.parse import urlparse
         from specify_cli.authentication.http import open_url as _open_url
 
         try:
-            parsed_src = urlparse(download_url)
+            urlparse(download_url).port
         except ValueError:
             console.print(f"[red]Error:[/red] Invalid URL: {_escape_markup(download_url)}")
             raise typer.Exit(1)
-        src_host = parsed_src.hostname or ""
-        src_loopback = src_host == "localhost"
-        if not src_loopback:
-            try:
-                src_loopback = ip_address(src_host).is_loopback
-            except ValueError:
-                # Host is not an IP literal (e.g., a DNS name); keep default non-loopback.
-                pass
-        if parsed_src.scheme != "https" and not (parsed_src.scheme == "http" and src_loopback):
+        if not is_https_or_localhost_http(download_url):
             console.print("[red]Error:[/red] Only HTTPS URLs are allowed, except HTTP for localhost.")
             raise typer.Exit(1)
 
@@ -1647,16 +1627,7 @@ def workflow_add(
                 redirect_validator=_reject_insecure_download_redirect,
             ) as resp:
                 final_url = resp.geturl()
-                final_parsed = urlparse(final_url)
-                final_host = final_parsed.hostname or ""
-                final_lb = final_host == "localhost"
-                if not final_lb:
-                    try:
-                        final_lb = ip_address(final_host).is_loopback
-                    except ValueError:
-                        # Redirect host is not an IP literal; keep loopback as determined above.
-                        pass
-                if final_parsed.scheme != "https" and not (final_parsed.scheme == "http" and final_lb):
+                if not is_https_or_localhost_http(final_url):
                     console.print(
                         f"[red]Error:[/red] URL redirected to non-HTTPS: {_escape_markup(final_url)}"
                     )
@@ -1714,7 +1685,7 @@ def workflow_add(
     # 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():
@@ -1788,27 +1759,17 @@ def _install_workflow_from_catalog(
         raise typer.Exit(1)
 
     # Validate URL scheme (HTTPS required, HTTP allowed for localhost only)
-    from ipaddress import ip_address
     from urllib.parse import urlparse
 
     try:
         parsed_url = urlparse(workflow_url)
-        url_host = parsed_url.hostname or ""
+        parsed_url.port
     except ValueError:
         console.print(
             f"[red]Error:[/red] Workflow '{safe_wf_id}' has a malformed install URL."
         )
         raise typer.Exit(1)
-    is_loopback = False
-    if url_host == "localhost":
-        is_loopback = True
-    else:
-        try:
-            is_loopback = ip_address(url_host).is_loopback
-        except ValueError:
-            # Host is not an IP literal (e.g., a regular hostname); treat as non-loopback.
-            pass
-    if parsed_url.scheme != "https" and not (parsed_url.scheme == "http" and is_loopback):
+    if not is_https_or_localhost_http(workflow_url):
         console.print(
             f"[red]Error:[/red] Workflow '{safe_wf_id}' has an invalid install URL. "
             "Only HTTPS URLs are allowed, except HTTP for localhost/loopback."
@@ -1862,16 +1823,7 @@ def _install_workflow_from_catalog(
         ) as response:
             # Validate final URL after redirects
             final_url = response.geturl()
-            final_parsed = urlparse(final_url)
-            final_host = final_parsed.hostname or ""
-            final_loopback = final_host == "localhost"
-            if not final_loopback:
-                try:
-                    final_loopback = ip_address(final_host).is_loopback
-                except ValueError:
-                    # Host is not an IP literal (e.g., a regular hostname); treat as non-loopback.
-                    pass
-            if final_parsed.scheme != "https" and not (final_parsed.scheme == "http" and final_loopback):
+            if not is_https_or_localhost_http(final_url):
                 _safe_discard_staged_workflow_file(staged_file, workflow_dir, existed_before)
                 console.print(
                     f"[red]Error:[/red] Workflow '{safe_wf_id}' redirected to non-HTTPS URL: {_escape_markup(final_url)}"
@@ -2694,28 +2646,17 @@ def workflow_step_add(
             )
             raise typer.Exit(1)
 
-    from urllib.parse import urlparse
     from specify_cli.authentication.http import open_url as _open_url
 
     def _safe_fetch(url: str) -> bytes:
-        parsed = urlparse(url)
-        is_localhost = parsed.hostname in ("localhost", "127.0.0.1", "::1")
-        if parsed.scheme != "https" and not (parsed.scheme == "http" and is_localhost):
+        if not is_https_or_localhost_http(url):
             raise ValueError(f"Refusing to fetch from non-HTTPS URL: {url}")
-        if not parsed.hostname:
-            raise ValueError(f"Refusing to fetch from URL with no hostname: {url}")
         with _open_url(
             url, timeout=30, redirect_validator=_reject_insecure_download_redirect
         ) as resp:
             final_url = resp.geturl()
-            final_parsed = urlparse(final_url)
-            final_is_localhost = final_parsed.hostname in ("localhost", "127.0.0.1", "::1")
-            if final_parsed.scheme != "https" and not (
-                final_parsed.scheme == "http" and final_is_localhost
-            ):
+            if not is_https_or_localhost_http(final_url):
                 raise ValueError(f"Redirect to non-HTTPS URL: {final_url}")
-            if not final_parsed.hostname:
-                raise ValueError(f"Redirect to URL with no hostname: {final_url}")
             return _read_response_within_limit(resp)
 
     _validate_step_id_or_exit(step_id)

src/specify_cli/workflows/catalog.py

@@ -894,7 +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()

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/merge.py

@@ -292,9 +292,21 @@ def _traverse_and_apply(
                         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":
+        # Insert after: higher-priority overlays land closer to the anchor
+        # (reversed merge order), but a single overlay's own inserts must keep
+        # their declared order — mirroring the forward insert_before loop above.
+        # Reversing the whole flat list would also flip an overlay's own edits,
+        # so group contiguous same-layer edits and reverse the GROUP order only.
+        after_groups: list[list[tuple[OverlayLayer, OverlayEdit]]] = []
+        for layer, edit in edits:
+            if edit.operation != "insert_after":
+                continue
+            if after_groups and after_groups[-1][0][0] is layer:
+                after_groups[-1].append((layer, edit))
+            else:
+                after_groups.append([(layer, edit)])
+        for group in reversed(after_groups):
+            for layer, edit in group:
                 new_step = copy.deepcopy(edit.step)
                 _record_sources_recursively(new_step, layer.source, sources)
                 result.append(new_step)

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

@@ -189,7 +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:

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

@@ -26,7 +26,7 @@ class GateStep(StepBase):
     later with ``specify workflow resume``.
 
     The user's choice is stored in ``output.choice``.  ``on_reject``
-    controls abort / skip behaviour.
+    controls abort / skip / retry behaviour.
     """
 
     type_key = "gate"
@@ -168,7 +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/init/__init__.py

@@ -59,7 +59,7 @@ class InitStep(StepBase):
         Extra options for the integration (e.g. ``"--skills"`` or
         ``"--commands-dir .myagent/cmds"``).
     ``script``
-        Script type, ``sh`` or ``ps``.
+        Script type, ``sh``, ``ps``, or ``py``.
     ``force``
         Merge/overwrite without confirmation when the directory is not
         empty.

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

@@ -138,7 +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:

templates/commands/constitution.md

@@ -14,6 +14,25 @@ $ARGUMENTS
 
 You **MUST** consider the user input before proceeding (if not empty).
 
+## Scope Guard
+
+This command's own work is limited to updating the project constitution and propagating
+constitution-driven changes to the dependent artifacts identified in this command.
+
+- Classify every part of the user input as either constitution content or a separate,
+  non-governance intent.
+- If the input includes feature implementation, code generation, refactoring, building, or
+  deployment requests, you **MUST NOT** execute them. Extract them as deferred intents instead.
+- You **MUST NOT** create, modify, or delete application source files, feature routes,
+  components, tests, deployment files, or other artifacts unrelated to the constitution
+  workflow and its required propagation.
+- If it is unclear whether an instruction is constitution content, ask for clarification before
+  making changes.
+- After completing the constitution update, include a `Next Actions` section for each deferred
+  intent. List the original intent and suggest the appropriate follow-up Spec Kit command, such
+  as `__SPECKIT_COMMAND_SPECIFY__`, without invoking it.
+- If there are no non-governance intents, omit the `Next Actions` section.
+
 ## Pre-Execution Checks
 
 **Check for extension hooks (before constitution update)**:
@@ -104,6 +123,7 @@ Follow this execution flow:
    - New version and bump rationale.
    - Any files flagged for manual follow-up.
    - Suggested commit message (e.g., `docs: amend constitution to vX.Y.Z (principle additions + governance update)`).
+   - A `Next Actions` section for any deferred non-governance intents.
 
 Formatting & Style Requirements:
 

templates/commands/specify.md

@@ -139,9 +139,9 @@ Given that feature description, do this:
     7. Identify Key Entities (if data involved)
     8. Return: SUCCESS (spec ready for planning)
 
-6. Write the specification to SPEC_FILE using the template structure, replacing placeholders with concrete details derived from the feature description (arguments) while preserving section order and headings.
+7. Write the specification to SPEC_FILE using the template structure, replacing placeholders with concrete details derived from the feature description (arguments) while preserving section order and headings.
 
-7. **Specification Quality Validation**: After writing the initial spec, validate it against quality criteria:
+8. **Specification Quality Validation**: After writing the initial spec, validate it against quality criteria:
 
    a. **Create Spec Quality Checklist**: Generate a checklist file at `SPECIFY_FEATURE_DIRECTORY/checklists/requirements.md` using the checklist template structure with these validation items:
 

tests/contract/test_catalog_schema.py

@@ -43,6 +43,42 @@ def test_builtin_default_stack_when_no_config(tmp_path: Path):
     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 = {

tests/contract/test_manifest_schema.py

@@ -124,3 +124,44 @@ 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)
+
+
+@pytest.mark.parametrize("bad", [[], "", 0, False, "extensions"])
+def test_non_mapping_provides_rejected_including_falsy(bad):
+    # `data.get("provides") or {}` coerced a FALSY non-mapping ([], '', 0, False)
+    # to {} before the type check, so a malformed manifest passed validation as
+    # a bundle that provides nothing. Only an absent/None value means "empty".
+    data = valid_manifest_dict()
+    data["provides"] = bad
+    with pytest.raises(BundlerError, match="'provides' must be a mapping when present"):
+        BundleManifest.from_dict(data)
+
+
+@pytest.mark.parametrize("bad", [[], "", 0, False, "speckit>=0.1"])
+def test_non_mapping_requires_rejected_including_falsy(bad):
+    # Same falsy-coercion hole for `requires`.
+    data = valid_manifest_dict()
+    data["requires"] = bad
+    with pytest.raises(BundlerError, match="'requires' must be a mapping when present"):
+        BundleManifest.from_dict(data)
+
+
+def test_absent_provides_and_requires_do_not_raise_mapping_error():
+    # Absent (None) optional mappings default to empty and must NOT trigger the
+    # "must be a mapping when present" guard — that is reserved for present
+    # non-mappings. (Structural completeness, e.g. requires.speckit_version, is
+    # a separate concern checked by structural_errors().)
+    data = valid_manifest_dict()
+    data.pop("provides", None)
+    data.pop("requires", None)
+    BundleManifest.from_dict(data)  # does not raise BundlerError

tests/contract/test_wheel_core_pack_scripts.py

@@ -0,0 +1,40 @@
+"""Contract tests for the script variants bundled into the wheel's core_pack.
+
+``specify init --script <type>`` installs from ``specify_cli/core_pack/scripts/``
+when the CLI runs from a wheel. Any script variant that lives in the repository
+must therefore be force-included at build time, otherwise the generated
+commands reference scripts the released package never ships (#3665).
+"""
+
+from __future__ import annotations
+
+import tomllib
+from pathlib import Path
+
+REPO_ROOT = Path(__file__).parents[2]
+
+
+def _force_include() -> dict[str, str]:
+    with (REPO_ROOT / "pyproject.toml").open("rb") as pyproject_file:
+        pyproject = tomllib.load(pyproject_file)
+    return pyproject["tool"]["hatch"]["build"]["targets"]["wheel"]["force-include"]
+
+
+def test_every_script_variant_is_bundled_into_core_pack():
+    force_include = _force_include()
+    variants = sorted(
+        path.name for path in (REPO_ROOT / "scripts").iterdir() if path.is_dir()
+    )
+
+    assert variants, "expected at least one script variant under scripts/"
+    for variant in variants:
+        assert force_include.get(f"scripts/{variant}") == (
+            f"specify_cli/core_pack/scripts/{variant}"
+        ), f"scripts/{variant} is missing from the wheel force-include list"
+
+
+def test_python_script_variant_is_bundled():
+    # Explicit regression guard for #3665: `--script py` shipped skills that
+    # invoked python3 .specify/scripts/python/*.py while the wheel bundled
+    # only the bash and PowerShell variants.
+    assert _force_include()["scripts/python"] == "specify_cli/core_pack/scripts/python"

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/git/test_git_extension_python_parity.py

@@ -515,6 +515,32 @@ class TestAutoCommitParity:
         assert p.stderr.strip() == b.stderr.strip()
         assert self._last_message(bash_proj) == self._last_message(py_proj) == "spec done"
 
+    def test_custom_message_with_trailing_whitespace_after_quote(self, tmp_path: Path):
+        """Trailing whitespace after a closing quote must not leave the quote
+        dangling in the commit message. A raw close-quote strip anchored to
+        end-of-string skips the quote when spaces follow it (``spec done"  ``);
+        trimming first (matching the PowerShell twin) yields a clean message and
+        keeps bash/python in parity."""
+        bash_proj, py_proj = _twin_projects(tmp_path)
+        config = (
+            "auto_commit:\n"
+            "  default: false\n"
+            "  after_specify:\n"
+            "    enabled: true\n"
+            '    message: "spec done"  \n'  # trailing spaces after the closing quote
+        )
+        for proj in (bash_proj, py_proj):
+            _write_config(proj, config)
+            self._dirty(proj)
+        b = _run_bash("auto-commit.sh", bash_proj, "after_specify")
+        p = _run_py("auto-commit", py_proj, "after_specify")
+        _assert_parity(b, p)
+        assert (
+            self._last_message(bash_proj)
+            == self._last_message(py_proj)
+            == "spec done"
+        )
+
     def test_default_true_applies_to_unlisted_event(self, tmp_path: Path):
         bash_proj, py_proj = _twin_projects(tmp_path)
         for proj in (bash_proj, py_proj):

tests/http_helpers.py

@@ -1,15 +1,46 @@
-"""HTTP test helpers shared by version-related CLI tests."""
+"""HTTP test helpers shared by CLI tests."""
 
+import io
 import json
+import urllib.request
 from unittest.mock import MagicMock
 
+import pytest
+
 
 def mock_urlopen_response(payload: dict) -> MagicMock:
     """Build a urlopen context-manager mock whose read returns JSON."""
     body = json.dumps(payload).encode("utf-8")
     resp = MagicMock()
-    resp.read.return_value = body
+    resp.read.side_effect = io.BytesIO(body).read
     cm = MagicMock()
     cm.__enter__.return_value = resp
     cm.__exit__.return_value = False
     return cm
+
+
+@pytest.fixture(autouse=True)
+def route_opener_open_through_urlopen(monkeypatch):
+    """Route build_opener().open through urllib.request.urlopen.
+
+    ``open_url(...)`` fetches via ``build_opener(...).open()``, which bypasses
+    ``urllib.request.urlopen`` — and with it the urlopen patches these test
+    modules are built on.
+    Delegating ``open()`` to urlopen at call time keeps those patches
+    effective; the redirect handler's own behavior is covered by
+    ``TestRedirectStripping`` in test_authentication.py.
+
+    Import this fixture into a test module to activate it there.
+    """
+
+    class _UrlopenDelegatingOpener:
+        def open(self, req, data=None, timeout=None):
+            if data is None:
+                return urllib.request.urlopen(req, timeout=timeout)
+            return urllib.request.urlopen(req, data=data, timeout=timeout)
+
+    monkeypatch.setattr(
+        urllib.request,
+        "build_opener",
+        lambda *handlers: _UrlopenDelegatingOpener(),
+    )

tests/integrations/test_base.py

@@ -236,6 +236,24 @@ class TestBuildCommandInvocation:
             == "/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."""

tests/integrations/test_integration_catalog.py

@@ -6,6 +6,8 @@ import os
 import pytest
 import yaml
 
+from tests.http_helpers import route_opener_open_through_urlopen  # noqa: F401
+
 from specify_cli.integrations.catalog import (
     IntegrationCatalog,
     IntegrationCatalogEntry,
@@ -13,9 +15,34 @@ 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 +278,48 @@ 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))

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_lingma.py

@@ -1,5 +1,7 @@
 """Tests for LingmaIntegration."""
 
+from specify_cli.integrations import get_integration
+
 from .test_integration_base_skills import SkillsIntegrationTests
 
 
@@ -8,3 +10,9 @@ class TestLingmaIntegration(SkillsIntegrationTests):
     FOLDER = ".lingma/"
     COMMANDS_SUBDIR = "skills"
     REGISTRAR_DIR = ".lingma/skills"
+
+    def test_multi_install_safe(self):
+        # Lingma writes only to its isolated, static root .lingma/skills,
+        # disjoint from every other integration, so it must be co-install safe
+        # (mirrors trae/zcode and the kiro-cli #3471 precedent).
+        assert get_integration(self.KEY).multi_install_safe is True

tests/integrations/test_integration_state.py

@@ -84,3 +84,31 @@ def test_write_integration_json_strips_integration_key(tmp_path):
     assert state["integration"] == "claude"
     assert state["default_integration"] == "claude"
     assert state["installed_integrations"] == ["claude"]
+
+
+def test_with_integration_setting_recomputes_separator_from_retained_options():
+    """Updating only script_type must not drop an options-dependent separator.
+
+    Copilot resolves the command-ref separator to '-' when '--skills' options
+    are stored and '.' otherwise. A second call that changes only script_type
+    (parsed_options=None, raw_options=None) retains the stored parsed_options,
+    so invoke_separator must stay '-', not be recomputed from the None argument.
+    """
+    from specify_cli.integrations import get_integration
+    from specify_cli.integration_runtime import with_integration_setting
+
+    copilot = get_integration("copilot")
+
+    settings = with_integration_setting(
+        {}, "copilot", copilot, parsed_options={"skills": True}
+    )
+    assert settings["copilot"]["invoke_separator"] == "-"
+
+    settings2 = with_integration_setting(
+        {"integration_settings": settings}, "copilot", copilot, script_type="ps"
+    )
+    # parsed_options are retained (only script_type changed) ...
+    assert settings2["copilot"]["parsed_options"] == {"skills": True}
+    assert settings2["copilot"]["script"] == "ps"
+    # ... so the separator must reflect them, not the (None) argument.
+    assert settings2["copilot"]["invoke_separator"] == "-"

tests/integrations/test_registry.py

@@ -28,6 +28,7 @@ ALL_INTEGRATION_KEYS = [
     "gemini", "tabnine",
     # Stage 5 — skills, generic & option-driven integrations
     "codex", "kimi", "agy", "zed", "generic",
+    "droid",
 ]
 
 

tests/self_upgrade_helpers.py

@@ -18,7 +18,7 @@ from specify_cli._version import (
     _verify_upgrade,
 )
 from tests.conftest import strip_ansi
-from tests.http_helpers import mock_urlopen_response
+from tests.http_helpers import mock_urlopen_response, route_opener_open_through_urlopen
 
 __all__ = (
     "SENTINEL_GH_TOKEN",
@@ -31,6 +31,7 @@ __all__ = (
     "_verify_upgrade",
     "mock_urlopen_response",
     "requires_posix",
+    "route_opener_open_through_urlopen",
     "runner",
     "strip_ansi",
 )

tests/test_agent_config_consistency.py

@@ -20,6 +20,7 @@ ISSUE_TEMPLATE_AGENT_KEYS = [
     "codex",
     "cursor-agent",
     "devin",
+    "droid",
     "firebender",
     "forge",
     "gemini",

tests/test_authentication.py

@@ -14,6 +14,7 @@ Covers:
 from __future__ import annotations
 
 import base64
+import io
 import json
 import os
 
@@ -515,6 +516,23 @@ class TestAzureDevOpsAuth:
         with patch("specify_cli.authentication.azure_devops.subprocess.run", side_effect=boom):
             assert AzureDevOpsAuth().resolve_token(entry) is None
 
+    @pytest.mark.parametrize("payload", [[], {"accessToken": None}, {"accessToken": 123}])
+    def test_resolve_token_azure_cli_unexpected_json_shape_returns_none(
+        self, payload
+    ):
+        from unittest.mock import MagicMock, patch
+
+        entry = AuthConfigEntry(
+            hosts=("dev.azure.com",), provider="azure-devops", auth="azure-cli",
+        )
+        result = MagicMock(returncode=0, stdout=json.dumps(payload))
+
+        with patch(
+            "specify_cli.authentication.azure_devops.subprocess.run",
+            return_value=result,
+        ):
+            assert AzureDevOpsAuth().resolve_token(entry) is None
+
     def test_resolve_token_azure_ad_success(self, monkeypatch):
         """azure-ad acquires token via OAuth2 client credentials."""
         from unittest.mock import patch, MagicMock
@@ -524,10 +542,15 @@ class TestAzureDevOpsAuth:
             tenant_id="tid", client_id="cid", client_secret_env="MY_SECRET",
         )
         mock_resp = MagicMock()
-        mock_resp.read.return_value = b'{"access_token": "ad-acquired-token"}'
+        mock_resp.read.side_effect = io.BytesIO(b'{"access_token": "ad-acquired-token"}').read
         mock_resp.__enter__ = lambda s: s
         mock_resp.__exit__ = MagicMock(return_value=False)
-        with patch("urllib.request.urlopen", return_value=mock_resp):
+        # The token request goes through a strict-redirect opener (so a 307/308
+        # cannot forward the client_secret body to a non-HTTPS host), not bare
+        # urlopen; patch the opener it builds.
+        mock_opener = MagicMock()
+        mock_opener.open.return_value = mock_resp
+        with patch("urllib.request.build_opener", return_value=mock_opener):
             assert AzureDevOpsAuth().resolve_token(entry) == "ad-acquired-token"
 
     def test_resolve_token_azure_ad_missing_secret_returns_none(self, monkeypatch):
@@ -542,14 +565,123 @@ class TestAzureDevOpsAuth:
     def test_resolve_token_azure_ad_network_error_returns_none(self, monkeypatch):
         """azure-ad returns None on network errors."""
         import urllib.error
-        from unittest.mock import patch
+        from unittest.mock import MagicMock, patch
         monkeypatch.setenv("MY_SECRET", "secret-value")
         entry = AuthConfigEntry(
             hosts=("dev.azure.com",), provider="azure-devops", auth="azure-ad",
             tenant_id="tid", client_id="cid", client_secret_env="MY_SECRET",
         )
-        with patch("urllib.request.urlopen",
-                    side_effect=urllib.error.URLError("connection refused")):
+        mock_opener = MagicMock()
+        mock_opener.open.side_effect = urllib.error.URLError("connection refused")
+        with patch("urllib.request.build_opener", return_value=mock_opener):
+            assert AzureDevOpsAuth().resolve_token(entry) is None
+
+    @pytest.mark.parametrize(
+        ("status", "reason"),
+        [(307, "Temporary Redirect"), (308, "Permanent Redirect")],
+    )
+    def test_resolve_token_azure_ad_rejects_https_redirect(
+        self, monkeypatch, status, reason
+    ):
+        """The client-secret POST must never be redirected to another host."""
+        import urllib.error
+        from unittest.mock import MagicMock, patch
+        from urllib.request import Request
+
+        monkeypatch.setenv("MY_SECRET", "secret-value")
+        entry = AuthConfigEntry(
+            hosts=("dev.azure.com",), provider="azure-devops", auth="azure-ad",
+            tenant_id="tid", client_id="cid", client_secret_env="MY_SECRET",
+        )
+        mock_opener = MagicMock()
+        mock_opener.open.side_effect = urllib.error.URLError("stop after setup")
+
+        with patch("urllib.request.build_opener", return_value=mock_opener) as build_opener:
+            assert AzureDevOpsAuth().resolve_token(entry) is None
+
+        redirect_handler = build_opener.call_args.args[0]
+        request = Request(
+            "https://login.microsoftonline.com/tid/oauth2/v2.0/token",
+            data=b"grant_type=client_credentials&client_secret=secret-value",
+            headers={"Content-Type": "application/x-www-form-urlencoded"},
+        )
+        assert request.get_method() == "POST"
+        assert b"client_secret=secret-value" in request.data
+        with pytest.raises(urllib.error.URLError, match="must not be redirected"):
+            redirect_handler.redirect_request(
+                request,
+                io.BytesIO(b""),
+                status,
+                reason,
+                {},
+                "https://evil.example/token",
+            )
+
+    def test_resolve_token_azure_ad_oversized_response_returns_none(
+        self, monkeypatch
+    ):
+        """Oversized token metadata is rejected before JSON parsing."""
+        from unittest.mock import MagicMock, patch
+
+        from specify_cli._download_security import MAX_JSON_METADATA_BYTES
+
+        monkeypatch.setenv("MY_SECRET", "secret-value")
+        entry = AuthConfigEntry(
+            hosts=("dev.azure.com",), provider="azure-devops", auth="azure-ad",
+            tenant_id="tid", client_id="cid", client_secret_env="MY_SECRET",
+        )
+        mock_resp = MagicMock()
+        mock_resp.read.side_effect = io.BytesIO(
+            b"x" * (MAX_JSON_METADATA_BYTES + 1)
+        ).read
+        mock_resp.__enter__ = lambda s: s
+        mock_resp.__exit__ = MagicMock(return_value=False)
+        mock_opener = MagicMock()
+        mock_opener.open.return_value = mock_resp
+
+        with patch("urllib.request.build_opener", return_value=mock_opener), patch(
+            "specify_cli.authentication.azure_devops._json.loads",
+            side_effect=AssertionError("oversized body must not be parsed"),
+        ):
+            assert AzureDevOpsAuth().resolve_token(entry) is None
+
+    @pytest.mark.parametrize("payload", [[], {"access_token": None}, {"access_token": 123}])
+    def test_resolve_token_azure_ad_unexpected_json_shape_returns_none(
+        self, monkeypatch, payload
+    ):
+        from unittest.mock import MagicMock, patch
+
+        monkeypatch.setenv("MY_SECRET", "secret-value")
+        entry = AuthConfigEntry(
+            hosts=("dev.azure.com",), provider="azure-devops", auth="azure-ad",
+            tenant_id="tid", client_id="cid", client_secret_env="MY_SECRET",
+        )
+        mock_resp = MagicMock()
+        mock_resp.read.side_effect = io.BytesIO(json.dumps(payload).encode()).read
+        mock_resp.__enter__ = lambda s: s
+        mock_resp.__exit__ = MagicMock(return_value=False)
+        mock_opener = MagicMock()
+        mock_opener.open.return_value = mock_resp
+
+        with patch("urllib.request.build_opener", return_value=mock_opener):
+            assert AzureDevOpsAuth().resolve_token(entry) is None
+
+    def test_resolve_token_azure_ad_invalid_utf8_returns_none(self, monkeypatch):
+        """azure-ad returns None when the token response is not valid UTF-8."""
+        from unittest.mock import MagicMock, patch
+        monkeypatch.setenv("MY_SECRET", "secret-value")
+        entry = AuthConfigEntry(
+            hosts=("dev.azure.com",), provider="azure-devops", auth="azure-ad",
+            tenant_id="tid", client_id="cid", client_secret_env="MY_SECRET",
+        )
+        mock_resp = MagicMock()
+        mock_resp.read.side_effect = io.BytesIO(b"\xff").read
+        mock_resp.__enter__ = lambda s: s
+        mock_resp.__exit__ = MagicMock(return_value=False)
+        mock_opener = MagicMock()
+        mock_opener.open.return_value = mock_resp
+
+        with patch("urllib.request.build_opener", return_value=mock_opener):
             assert AzureDevOpsAuth().resolve_token(entry) is None
 
 
@@ -615,13 +747,15 @@ class TestAuthenticatedHttp:
         monkeypatch.setenv("GH_TOKEN", "my-token")
         self._set_config(monkeypatch, [_github_entry()])
         captured = {}
-        def fake_urlopen(req, timeout=None):
+        def fake_open(req, timeout=None):
             captured["req"] = req
             resp = MagicMock()
             resp.__enter__ = lambda s: s
             resp.__exit__ = MagicMock(return_value=False)
             return resp
-        with patch("specify_cli.authentication.http.urllib.request.urlopen", side_effect=fake_urlopen):
+        mock_opener = MagicMock()
+        mock_opener.open.side_effect = fake_open
+        with patch("specify_cli.authentication.http.urllib.request.build_opener", return_value=mock_opener):
             open_url("https://example.com/file.json")
         assert captured["req"].get_header("Authorization") is None
 
@@ -630,13 +764,15 @@ class TestAuthenticatedHttp:
         from specify_cli.authentication.http import open_url
         self._set_config(monkeypatch, [])
         captured = {}
-        def fake_urlopen(req, timeout=None):
+        def fake_open(req, timeout=None):
             captured["req"] = req
             resp = MagicMock()
             resp.__enter__ = lambda s: s
             resp.__exit__ = MagicMock(return_value=False)
             return resp
-        with patch("specify_cli.authentication.http.urllib.request.urlopen", side_effect=fake_urlopen):
+        mock_opener = MagicMock()
+        mock_opener.open.side_effect = fake_open
+        with patch("specify_cli.authentication.http.urllib.request.build_opener", return_value=mock_opener):
             open_url("https://github.com/org/repo")
         assert captured["req"].get_header("Authorization") is None
 
@@ -658,8 +794,7 @@ class TestAuthenticatedHttp:
             return resp
         mock_opener = MagicMock()
         mock_opener.open.side_effect = fake_side_effect
-        with patch("specify_cli.authentication.http.urllib.request.build_opener", return_value=mock_opener), \
-             patch("specify_cli.authentication.http.urllib.request.urlopen", side_effect=fake_side_effect):
+        with patch("specify_cli.authentication.http.urllib.request.build_opener", return_value=mock_opener):
             open_url("https://github.com/org/repo")
         assert call_count == 2
 
@@ -700,21 +835,23 @@ class TestAuthenticatedHttpNegative:
 
     def test_urlerror_propagates(self, monkeypatch):
         import urllib.error
-        from unittest.mock import patch
+        from unittest.mock import MagicMock, patch
         from specify_cli.authentication.http import open_url
         self._set_config(monkeypatch, [])
-        with patch("specify_cli.authentication.http.urllib.request.urlopen",
-                    side_effect=urllib.error.URLError("refused")):
+        mock_opener = MagicMock()
+        mock_opener.open.side_effect = urllib.error.URLError("refused")
+        with patch("specify_cli.authentication.http.urllib.request.build_opener", return_value=mock_opener):
             with pytest.raises(urllib.error.URLError):
                 open_url("https://example.com/file")
 
     def test_timeout_propagates(self, monkeypatch):
         import socket
-        from unittest.mock import patch
+        from unittest.mock import MagicMock, patch
         from specify_cli.authentication.http import open_url
         self._set_config(monkeypatch, [])
-        with patch("specify_cli.authentication.http.urllib.request.urlopen",
-                    side_effect=socket.timeout("timed out")):
+        mock_opener = MagicMock()
+        mock_opener.open.side_effect = socket.timeout("timed out")
+        with patch("specify_cli.authentication.http.urllib.request.build_opener", return_value=mock_opener):
             with pytest.raises(socket.timeout):
                 open_url("https://example.com/file")
 
@@ -820,17 +957,18 @@ class TestRedirectStripping:
         assert new_req.headers.get("Authorization") is None
         assert new_req.unredirected_hdrs.get("Authorization") is None
 
-    def test_https_to_http_same_host_redirect_strips_auth(self):
+    def test_https_to_http_same_host_redirect_rejected(self):
         from specify_cli.authentication.http import _StripAuthOnRedirect
         from urllib.request import Request
         import io
+        import urllib.error
+
         handler = _StripAuthOnRedirect(("github.com",))
         req = Request("https://github.com/org/repo", headers={"Authorization": "Bearer tok"})
-        new_req = handler.redirect_request(req, io.BytesIO(b""), 302, "Found", {},
-                                           "http://github.com/org/repo")
-        assert new_req is not None
-        assert new_req.headers.get("Authorization") is None
-        assert new_req.unredirected_hdrs.get("Authorization") is None
+
+        with pytest.raises(urllib.error.URLError, match="unsafe redirect"):
+            handler.redirect_request(req, io.BytesIO(b""), 302, "Found", {},
+                                     "http://github.com/org/repo")
 
     def test_redirect_validator_can_reject_before_following_redirect(self):
         import urllib.error
@@ -888,6 +1026,177 @@ class TestRedirectStripping:
             handler.redirect_request(req, io.BytesIO(b""), 302, "Found", {},
                                      "https://[::1/asset")
 
+    def test_redirect_rejects_https_downgrade(self):
+        """HTTPS downloads must not follow redirects to non-local HTTP URLs."""
+        from specify_cli.authentication.http import _StripAuthOnRedirect
+        from urllib.request import Request
+        import io
+        import urllib.error
+        handler = _StripAuthOnRedirect(("example.com",))
+        req = Request("https://example.com/archive.zip")
+        with pytest.raises(urllib.error.URLError, match="unsafe redirect"):
+            handler.redirect_request(req, io.BytesIO(b""), 302, "Found", {},
+                                     "http://evil.example.com/archive.zip")
+
+    @pytest.mark.parametrize(
+        "target",
+        [
+            "http://127.0.0.1/internal",
+            "https://localhost/internal",
+            "https://localhost./internal",
+            "https://service.localhost/internal",
+            "https://service.localhost./internal",
+            "https://127.0.0.2/internal",
+            "https://127.1/internal",
+            "https://2130706433/internal",
+            "https://0x7f000001/internal",
+            "https://017700000001/internal",
+            "https://0177.0.0.1/internal",
+            "https://[::1]/internal",
+            "https://[::1%25lo0]/internal",
+            "https://[::ffff:127.0.0.1]/internal",
+            "https://127%2e0%2e0%2e1/internal",
+            "https://%31%32%37.0.0.1/internal",
+            "https://127%2E1/internal",
+            "https://local%68ost/internal",
+            "https://[::ffff:127%2e0.0.1]/internal",
+            "https://[::ffff:7f00%3a1]/internal",
+            "https://[::ffff%3a127.0.0.1]/internal",
+            "https://ℓocalhost/internal",
+            "https://ｌｏｃａｌｈｏｓｔ/internal",
+            "https://127。0。0。1/internal",
+            "https://0.0.0.0/internal",
+            "https://0/internal",
+            "https://00.00.00.00/internal",
+            "https://[::]/internal",
+            "https://[::ffff:0.0.0.0]/internal",
+        ],
+    )
+    def test_redirect_rejects_remote_to_loopback(self, target):
+        """A remote response must not redirect a download into loopback."""
+        import io
+        import urllib.error
+        from urllib.request import Request
+
+        from specify_cli.authentication.http import _StripAuthOnRedirect
+
+        handler = _StripAuthOnRedirect(())
+        request = Request("https://example.com/archive.zip")
+
+        with pytest.raises(urllib.error.URLError, match="unsafe redirect"):
+            handler.redirect_request(
+                request,
+                io.BytesIO(b""),
+                302,
+                "Found",
+                {},
+                target,
+            )
+
+    @pytest.mark.parametrize(
+        ("source", "target"),
+        [
+            (
+                "http://localhost:8000/archive.zip",
+                "http://127.0.0.1:8001/archive.zip",
+            ),
+            (
+                "http://127.0.0.2:8000/archive.zip",
+                "http://127.255.255.254:8001/archive.zip",
+            ),
+            (
+                "https://[0:0:0:0:0:0:0:1]/archive.zip",
+                "http://[::1]:8001/archive.zip",
+            ),
+        ],
+    )
+    def test_redirect_allows_loopback_to_http_loopback(self, source, target):
+        """Local development may continue redirecting between loopback URLs."""
+        import io
+        from urllib.request import Request
+
+        from specify_cli.authentication.http import _StripAuthOnRedirect
+
+        handler = _StripAuthOnRedirect(())
+        request = Request(source)
+        redirected = handler.redirect_request(
+            request,
+            io.BytesIO(b""),
+            302,
+            "Found",
+            {},
+            target,
+        )
+
+        assert redirected is not None
+
+    def test_multi_hop_remote_to_loopback_chain_is_rejected_at_first_hop(self):
+        import io
+        import urllib.error
+        from urllib.request import Request
+
+        from specify_cli.authentication.http import _StripAuthOnRedirect
+
+        handler = _StripAuthOnRedirect(())
+        request = Request("https://example.com/archive.zip")
+
+        with pytest.raises(urllib.error.URLError, match="unsafe redirect"):
+            handler.redirect_request(
+                request,
+                io.BytesIO(b""),
+                302,
+                "Found",
+                {},
+                "https://localhost:4443/hop",
+            )
+
+    @pytest.mark.parametrize(
+        "target",
+        [
+            "https://example.com:notaport/archive.zip",
+            "https://example.com:+443/archive.zip",
+            "https://example.com:65536/archive.zip",
+        ],
+    )
+    def test_malformed_redirect_port_raises_urlerror(self, target):
+        import io
+        import urllib.error
+        from urllib.request import Request
+
+        from specify_cli.authentication.http import _StripAuthOnRedirect
+
+        handler = _StripAuthOnRedirect(())
+        request = Request("https://example.com/archive.zip")
+
+        with pytest.raises(urllib.error.URLError, match="malformed redirect URL"):
+            handler.redirect_request(
+                request,
+                io.BytesIO(b""),
+                302,
+                "Found",
+                {},
+                target,
+            )
+
+    def test_strict_redirect_error_describes_target_and_allowed_localhost(self):
+        from specify_cli.authentication.http import _StripAuthOnRedirect
+        from urllib.request import Request
+        import io
+        import urllib.error
+
+        handler = _StripAuthOnRedirect(("example.com",))
+        req = Request("https://example.com/archive.zip")
+
+        with pytest.raises(urllib.error.URLError) as exc_info:
+            handler.redirect_request(req, io.BytesIO(b""), 302, "Found", {},
+                                     "http://evil.example.com/archive.zip")
+
+        error_message = str(exc_info.value)
+        assert "http://evil.example.com/archive.zip" in error_message
+        assert "localhost" in error_message
+        assert "127.0.0.1" in error_message
+        assert "::1" in error_message
+
 
 # ---------------------------------------------------------------------------
 # _fetch_latest_release_tag delegation
@@ -907,7 +1216,7 @@ class TestFetchLatestReleaseTagDelegation:
             captured["request"] = req
             body = _json.dumps({"tag_name": "v9.9.9"}).encode()
             resp = MagicMock()
-            resp.read.return_value = body
+            resp.read.side_effect = io.BytesIO(body).read
             cm = MagicMock()
             cm.__enter__.return_value = resp
             cm.__exit__.return_value = False
@@ -927,20 +1236,25 @@ class TestFetchLatestReleaseTagDelegation:
         assert captured["request"].get_header("Authorization") == "Bearer forwarded-sentinel"
 
     def test_no_config_means_no_auth(self, monkeypatch):
-        from unittest.mock import patch
+        from unittest.mock import MagicMock, patch
         from specify_cli._version import _fetch_latest_release_tag
         self._set_config(monkeypatch, [])
         captured, side_effect = self._capture_request()
-        with patch("specify_cli.authentication.http.urllib.request.urlopen", side_effect=side_effect):
+        # The unauthenticated path uses the strict redirect opener too.
+        mock_opener = MagicMock()
+        mock_opener.open.side_effect = side_effect
+        with patch("specify_cli.authentication.http.urllib.request.build_opener", return_value=mock_opener):
             _fetch_latest_release_tag()
         assert captured["request"].get_header("Authorization") is None
 
     def test_accept_header_present(self, monkeypatch):
-        from unittest.mock import patch
+        from unittest.mock import MagicMock, patch
         from specify_cli._version import _fetch_latest_release_tag
         self._set_config(monkeypatch, [])
         captured, side_effect = self._capture_request()
-        with patch("specify_cli.authentication.http.urllib.request.urlopen", side_effect=side_effect):
+        mock_opener = MagicMock()
+        mock_opener.open.side_effect = side_effect
+        with patch("specify_cli.authentication.http.urllib.request.build_opener", return_value=mock_opener):
             _fetch_latest_release_tag()
         assert captured["request"].get_header("Accept") == "application/vnd.github+json"
 

tests/test_download_security.py

@@ -0,0 +1,227 @@
+"""Tests for bounded HTTP download helpers."""
+
+from __future__ import annotations
+
+import weakref
+
+import pytest
+
+from specify_cli._download_security import (
+    is_https_or_localhost_http,
+    is_loopback_url,
+    read_response_limited,
+)
+
+
+@pytest.mark.parametrize(
+    "url, allowed",
+    [
+        ("https://example.com/preset.zip", True),
+        ("http://localhost:8000/preset.zip", True),
+        ("http://127.0.0.1/preset.zip", True),
+        ("http://127.0.0.2/preset.zip", True),
+        ("http://127.255.255.254/preset.zip", True),
+        ("http://[::1]/preset.zip", True),
+        ("http://[0:0:0:0:0:0:0:1]/preset.zip", True),
+        ("http://[::ffff:127.0.0.2]/preset.zip", True),
+        ("http://[::1%25lo0]/preset.zip", True),
+        # Non-loopback HTTP is rejected.
+        ("http://example.com/preset.zip", False),
+        ("http://192.0.2.1/preset.zip", False),
+        ("http://[fe80::1]/preset.zip", False),
+        ("http://[fe80::1%25lo0]/preset.zip", False),
+        ("http://0.0.0.0/preset.zip", False),
+        ("http://0/preset.zip", False),
+        ("http://[::]/preset.zip", False),
+        ("http://[::ffff:0.0.0.0]/preset.zip", False),
+        # Ambiguous/platform-dependent spellings may never authorize HTTP.
+        ("http://127.1/preset.zip", False),
+        ("http://2130706433/preset.zip", False),
+        ("http://0x7f000001/preset.zip", False),
+        ("http://017700000001/preset.zip", False),
+        ("http://0177.0.0.1/preset.zip", False),
+        ("http://00177.0.0.1/preset.zip", False),
+        ("http://localhost./preset.zip", False),
+        ("http://ℓocalhost/preset.zip", False),
+        ("http://127。0。0。1/preset.zip", False),
+        # A hostname is always required, even for HTTPS.
+        ("https:///preset.zip", False),
+        ("https://", False),
+        # Invalid ports must be rejected before urllib opens the URL.
+        ("https://example.com:notaport/preset.zip", False),
+        ("https://example.com:+443/preset.zip", False),
+        ("https://example.com:65536/preset.zip", False),
+        # urllib decodes escapes in the authority before connecting; reject
+        # encoded reg-names so validation and connection cannot disagree.
+        ("https://127%2e0%2e0%2e1/preset.zip", False),
+        ("https://%31%32%37.0.0.1/preset.zip", False),
+        ("https://local%68ost/preset.zip", False),
+        ("https://example.com%3a443/preset.zip", False),
+        ("https://[::1%lo0]/preset.zip", False),
+        ("https://[::ffff:127%2e0.0.1]/preset.zip", False),
+        ("https://[::ffff:7f00%3a1]/preset.zip", False),
+        ("https://[::ffff%3a127.0.0.1]/preset.zip", False),
+    ],
+)
+def test_is_https_or_localhost_http(url, allowed):
+    assert is_https_or_localhost_http(url) is allowed
+
+
+@pytest.mark.parametrize(
+    "url",
+    [
+        "https://localhost/internal",
+        "https://127.0.0.2/internal",
+        "https://[::1]/internal",
+        "https://[::1%25lo0]/internal",
+        "https://[::ffff:127.0.0.2]/internal",
+    ],
+)
+def test_is_loopback_url_recognizes_effective_loopback_literals(url):
+    assert is_loopback_url(url) is True
+
+
+@pytest.mark.parametrize(
+    "url",
+    [
+        "https://localhost./internal",
+        "https://service.localhost/internal",
+        "https://service.localhost./internal",
+        "https://127.1/internal",
+        "https://2130706433/internal",
+        "https://0x7f000001/internal",
+        "https://017700000001/internal",
+        "https://0177.0.0.1/internal",
+        "https://ℓocalhost/internal",
+        "https://127。0。0。1/internal",
+        "https://127%2e0%2e0%2e1/internal",
+        "https://0.0.0.0/internal",
+        "https://0/internal",
+        "https://00.00.00.00/internal",
+        "https://[::]/internal",
+        "https://[::ffff:0.0.0.0]/internal",
+    ],
+)
+def test_is_loopback_url_does_not_authorize_ambiguous_spellings(url):
+    assert is_loopback_url(url) is False
+
+
+class _Response:
+    """Faithful stream stand-in: read() advances a cursor and returns b"" at EOF."""
+
+    def __init__(self, data: bytes, *, chunk: int | None = None):
+        self.data = data
+        self.pos = 0
+        # When set, never return more than *chunk* bytes per call even if more is
+        # requested - simulates short reads (e.g. chunked transfer encoding).
+        self.chunk = chunk
+
+    def read(self, size: int = -1) -> bytes:
+        if size < 0:
+            size = len(self.data) - self.pos
+        if self.chunk is not None:
+            size = min(size, self.chunk)
+        out = self.data[self.pos : self.pos + size]
+        self.pos += len(out)
+        return out
+
+
+class _RecordingResponse(_Response):
+    def __init__(self, data: bytes, *, chunk: int | None = None):
+        super().__init__(data, chunk=chunk)
+        self.requested_sizes: list[int] = []
+
+    def read(self, size: int = -1) -> bytes:
+        self.requested_sizes.append(size)
+        return super().read(size)
+
+
+class _TrackedChunk(bytearray):
+    pass
+
+
+class _OneByteResponse:
+    """Return distinct weak-referenceable chunks to detect retained fragments."""
+
+    def __init__(self, count: int):
+        self.remaining = count
+        self.refs: list[weakref.ReferenceType[_TrackedChunk]] = []
+        self.peak_live = 0
+
+    def read(self, _size: int = -1) -> bytes | _TrackedChunk:
+        if self.remaining == 0:
+            return b""
+        self.remaining -= 1
+        chunk = _TrackedChunk(b"x")
+        self.refs.append(weakref.ref(chunk))
+        self.peak_live = max(
+            self.peak_live,
+            sum(ref() is not None for ref in self.refs),
+        )
+        return chunk
+
+
+def test_read_response_limited_rejects_oversized_download():
+    with pytest.raises(ValueError, match="exceeds maximum size"):
+        read_response_limited(_Response(b"abcde"), max_bytes=4)
+
+
+def test_read_response_limited_returns_full_body_within_limit():
+    assert read_response_limited(_Response(b"abcde"), max_bytes=10) == b"abcde"
+
+
+def test_read_response_limited_enforces_bound_under_short_reads():
+    # A server that streams more than max_bytes total while every read() returns
+    # fewer bytes than requested (chunked encoding) must still be rejected - a
+    # single read(max_bytes + 1) could be fooled, the accumulating loop cannot.
+    response = _Response(b"x" * 100, chunk=8)
+    with pytest.raises(ValueError, match="exceeds maximum size"):
+        read_response_limited(response, max_bytes=16)
+
+
+def test_read_response_limited_does_not_retain_short_read_fragments():
+    response = _OneByteResponse(64)
+
+    assert read_response_limited(response, max_bytes=64) == b"x" * 64
+    assert response.peak_live <= 2
+
+
+def test_read_response_limited_caps_underlying_reads_at_64_kib():
+    response = _RecordingResponse(b"x" * (64 * 1024 + 1))
+
+    with pytest.raises(ValueError, match="exceeds maximum size"):
+        read_response_limited(response, max_bytes=64 * 1024)
+
+    assert max(response.requested_sizes) <= 64 * 1024
+
+
+@pytest.mark.parametrize("value", [None, "1", 1.5, True])
+def test_read_response_limited_rejects_non_integer_limits(value):
+    with pytest.raises(TypeError, match="integer"):
+        read_response_limited(_Response(b""), max_bytes=value)
+
+
+def test_read_response_limited_rejects_negative_limit_without_reading():
+    response = _RecordingResponse(b"")
+
+    with pytest.raises(ValueError, match="non-negative"):
+        read_response_limited(response, max_bytes=-1)
+
+    assert response.requested_sizes == []
+
+
+def test_read_response_limited_allows_empty_response_at_zero_limit():
+    assert read_response_limited(_Response(b""), max_bytes=0) == b""
+
+
+class _CustomLimitError(Exception):
+    pass
+
+
+def test_read_response_limited_rejects_first_byte_at_zero_limit():
+    with pytest.raises(_CustomLimitError, match="exceeds maximum size"):
+        read_response_limited(
+            _Response(b"x"),
+            max_bytes=0,
+            error_type=_CustomLimitError,
+        )

tests/test_extensions.py

@@ -9,6 +9,7 @@ Tests cover:
 - Catalog stack (multi-catalog support)
 """
 
+import io
 import pytest
 import json
 import os
@@ -22,6 +23,7 @@ from datetime import datetime, timezone
 from unittest.mock import MagicMock
 
 from tests.conftest import strip_ansi
+from tests.http_helpers import route_opener_open_through_urlopen  # noqa: F401
 from specify_cli import extensions as _ext_module
 from specify_cli.extensions import (
     CatalogEntry,
@@ -4978,7 +4980,7 @@ class TestExtensionCatalog:
         zip_bytes = zip_buf.getvalue()
 
         release_response = MagicMock()
-        release_response.read.return_value = json.dumps(
+        release_response.read.side_effect = io.BytesIO(json.dumps(
             {
                 "assets": [
                     {
@@ -4987,12 +4989,12 @@ class TestExtensionCatalog:
                     }
                 ]
             }
-        ).encode()
+        ).encode()).read
         release_response.__enter__ = lambda s: s
         release_response.__exit__ = MagicMock(return_value=False)
 
         asset_response = MagicMock()
-        asset_response.read.return_value = zip_bytes
+        asset_response.read.side_effect = io.BytesIO(zip_bytes).read
         asset_response.__enter__ = lambda s: s
         asset_response.__exit__ = MagicMock(return_value=False)
 
@@ -8238,6 +8240,42 @@ class TestHookInvocationRendering:
         assert execution["command"] == "my-extension.do-something"
         assert execution["invocation"] == "/speckit-my-extension-do-something"
 
+    def test_forge_hooks_render_hyphenated_invocation(self, project_dir):
+        """Forge projects should render /speckit-* invocations (like Cline)."""
+        init_options = project_dir / ".specify" / "init-options.json"
+        init_options.parent.mkdir(parents=True, exist_ok=True)
+        init_options.write_text(json.dumps({"ai": "forge"}))
+
+        hook_executor = HookExecutor(project_dir)
+        execution = hook_executor.execute_hook(
+            {
+                "extension": "test-ext",
+                "command": "speckit.tasks",
+                "optional": False,
+            }
+        )
+
+        assert execution["command"] == "speckit.tasks"
+        assert execution["invocation"] == "/speckit-tasks"
+
+    def test_forge_hooks_render_extension_command(self, project_dir):
+        """Forge projects should render /speckit-my-ext-cmd for extension hooks."""
+        init_options = project_dir / ".specify" / "init-options.json"
+        init_options.parent.mkdir(parents=True, exist_ok=True)
+        init_options.write_text(json.dumps({"ai": "forge"}))
+
+        hook_executor = HookExecutor(project_dir)
+        execution = hook_executor.execute_hook(
+            {
+                "extension": "test-ext",
+                "command": "my-extension.do-something",
+                "optional": False,
+            }
+        )
+
+        assert execution["command"] == "my-extension.do-something"
+        assert execution["invocation"] == "/speckit-my-extension-do-something"
+
     def test_non_skill_command_keeps_slash_invocation(self, project_dir):
         """Custom hook commands should keep slash invocation style."""
         init_options = project_dir / ".specify" / "init-options.json"
@@ -8725,10 +8763,10 @@ def test_extension_wrapper_resolves_ghes_asset_when_host_configured(tmp_path, mo
     def fake_open(url, timeout=None, extra_headers=None):
         captured.append(url)
         resp = MagicMock()
-        resp.read.return_value = json.dumps({
+        resp.read.side_effect = io.BytesIO(json.dumps({
             "assets": [{"name": "ext.zip",
                         "url": "https://ghes.example/api/v3/repos/o/r/releases/assets/7"}]
-        }).encode()
+        }).encode()).read
         yield resp
 
     monkeypatch.setattr(catalog, "_open_url", fake_open)

tests/test_github_http.py

@@ -1,16 +1,20 @@
 """Tests for GitHub-authenticated HTTP request helpers."""
 
+import io
 import json
 import os
 from contextlib import contextmanager
 from unittest.mock import MagicMock, patch
+from urllib.request import Request
 
 import pytest
 
 from specify_cli._github_http import (
+    GITHUB_HOSTS,
     build_github_request,
     resolve_github_release_asset_api_url,
 )
+from specify_cli.authentication.http import _StripAuthOnRedirect
 
 
 class TestBuildGitHubRequest:
@@ -90,7 +94,7 @@ class TestResolveGitHubReleaseAssetApiUrl:
         @contextmanager
         def fake_open(url, timeout=None, extra_headers=None):
             resp = MagicMock()
-            resp.read.return_value = json.dumps(release_json).encode()
+            resp.read.side_effect = io.BytesIO(json.dumps(release_json).encode()).read
             yield resp
         return fake_open
 
@@ -198,7 +202,7 @@ class TestResolveGitHubReleaseAssetApiUrl:
         def capturing_open(url, timeout=None, extra_headers=None):
             captured_urls.append(url)
             resp = MagicMock()
-            resp.read.return_value = json.dumps({"assets": []}).encode()
+            resp.read.side_effect = io.BytesIO(json.dumps({"assets": []}).encode()).read
             yield resp
 
         resolve_github_release_asset_api_url(
@@ -217,7 +221,7 @@ class TestResolveGitHubReleaseAssetApiUrl:
         def capturing_open(url, timeout=None, extra_headers=None):
             captured_urls.append(url)
             resp = MagicMock()
-            resp.read.return_value = json.dumps({"assets": []}).encode()
+            resp.read.side_effect = io.BytesIO(json.dumps({"assets": []}).encode()).read
             yield resp
 
         resolve_github_release_asset_api_url(
@@ -260,7 +264,7 @@ class TestResolveGitHubReleaseAssetApiUrl:
         def recording_open(url, timeout=None, extra_headers=None):
             called.append(url)
             resp = MagicMock()
-            resp.read.return_value = b"{}"
+            resp.read.side_effect = io.BytesIO(b"{}").read
             yield resp
 
         result = resolve_github_release_asset_api_url(
@@ -299,7 +303,7 @@ class TestResolveGitHubReleaseAssetApiUrl:
         def recording_open(url, timeout=None, extra_headers=None):
             called.append(url)
             resp = MagicMock()
-            resp.read.return_value = b"{}"
+            resp.read.side_effect = io.BytesIO(b"{}").read
             yield resp
 
         url = "https://ghes.example/api/v3/repos/o/r/releases/assets/7"
@@ -317,7 +321,7 @@ class TestResolveGitHubReleaseAssetApiUrl:
         def capturing_open(url, timeout=None, extra_headers=None):
             captured.append(url)
             resp = MagicMock()
-            resp.read.return_value = json.dumps({"assets": []}).encode()
+            resp.read.side_effect = io.BytesIO(json.dumps({"assets": []}).encode()).read
             yield resp
 
         resolve_github_release_asset_api_url(
@@ -344,10 +348,10 @@ class TestResolveGitHubReleaseAssetApiUrl:
         def capturing_open(url, timeout=None, extra_headers=None):
             captured.append(url)
             resp = MagicMock()
-            resp.read.return_value = json.dumps({
+            resp.read.side_effect = io.BytesIO(json.dumps({
                 "assets": [{"name": "pack.zip",
                             "url": "https://api.github.com/repos/org/repo/releases/assets/99"}]
-            }).encode()
+            }).encode()).read
             yield resp
 
         result = resolve_github_release_asset_api_url(
@@ -357,3 +361,43 @@ class TestResolveGitHubReleaseAssetApiUrl:
         )
         assert result == "https://api.github.com/repos/org/repo/releases/assets/99"
         assert captured == ["https://api.github.com/repos/org/repo/releases/tags/v1.0"]
+
+
+class TestGitHubRedirectAuth:
+    """Tests for GitHub-owned redirect auth handling."""
+
+    def test_multi_hop_github_redirect_preserves_unredirected_auth(self):
+        """Auth survives a multi-hop redirect chain within GitHub hosts."""
+        handler = _StripAuthOnRedirect(tuple(GITHUB_HOSTS))
+        req1 = Request(
+            "https://github.com/org/repo",
+            headers={"Authorization": "Bearer tok"},
+        )
+
+        req2 = handler.redirect_request(
+            req1,
+            io.BytesIO(b""),
+            302,
+            "Found",
+            {},
+            "https://codeload.github.com/org/repo/zip",
+        )
+        assert req2 is not None
+        auth2 = req2.get_header("Authorization") or req2.unredirected_hdrs.get(
+            "Authorization"
+        )
+        assert auth2 == "Bearer tok"
+
+        req3 = handler.redirect_request(
+            req2,
+            io.BytesIO(b""),
+            302,
+            "Found",
+            {},
+            "https://raw.githubusercontent.com/org/repo/main/file",
+        )
+        assert req3 is not None
+        auth3 = req3.get_header("Authorization") or req3.unredirected_hdrs.get(
+            "Authorization"
+        )
+        assert auth3 == "Bearer tok"

tests/test_post_process.py

@@ -243,3 +243,34 @@ class TestRegressionPlainTemplate:
         assert output_file.exists(), f"Output file missing for {agent}"
         content = output_file.read_text(encoding="utf-8")
         assert body_text.strip() in content, f"Body text missing in {agent} output"
+
+
+class TestClineRealPostProcess:
+    """Cline's real command-content transforms (hook-command note + handoff
+    dot->hyphen rewrite) must run for extension/preset commands registered via
+    CommandRegistrar. This exercises the REAL method (not a monkeypatched
+    marker), so it fails if Cline's override does not match the base hook name
+    the registrar dispatches to (post_process_command_content)."""
+
+    def test_cline_transforms_applied_via_registrar(
+        self, tmp_path, registrar, ext_dir
+    ):
+        ext, cmd_dir = ext_dir
+        body = (
+            "- For each executable hook, output the following:\n"
+            "agent: speckit.foo\n"
+        )
+        _write_cmd(cmd_dir, body=body)
+
+        commands = [{"name": "speckit.test.review", "file": "commands/review.md"}]
+        registrar.register_commands("cline", commands, "test-ext", ext, tmp_path)
+
+        outputs = list((tmp_path / ".clinerules" / "workflows").rglob("*.md"))
+        assert outputs, "no cline command file was written"
+        content = outputs[0].read_text(encoding="utf-8")
+
+        # _inject_hook_command_note fired (its note text contains "replace dots")
+        assert "replace dots" in content
+        # _rewrite_handoff_references rewrote the dotted agent handoff
+        assert "agent: speckit-foo" in content
+        assert "agent: speckit.foo" not in content

tests/test_presets.py

@@ -2303,7 +2303,7 @@ class TestPresetCatalog:
         zip_bytes = zip_buf.getvalue()
 
         release_response = MagicMock()
-        release_response.read.return_value = json.dumps(
+        release_response.read.side_effect = io.BytesIO(json.dumps(
             {
                 "assets": [
                     {
@@ -2312,12 +2312,12 @@ class TestPresetCatalog:
                     }
                 ]
             }
-        ).encode()
+        ).encode()).read
         release_response.__enter__ = lambda s: s
         release_response.__exit__ = MagicMock(return_value=False)
 
         asset_response = MagicMock()
-        asset_response.read.return_value = zip_bytes
+        asset_response.read.side_effect = io.BytesIO(zip_bytes).read
         asset_response.__enter__ = lambda s: s
         asset_response.__exit__ = MagicMock(return_value=False)
 
@@ -5381,6 +5381,9 @@ class TestPresetEnableDisable:
 
 
 LEAN_PRESET_DIR = Path(__file__).parent.parent / "presets" / "lean"
+CORE_CONSTITUTION_COMMAND = (
+    Path(__file__).parent.parent / "templates" / "commands" / "constitution.md"
+)
 
 LEAN_COMMAND_NAMES = [
     "speckit.specify",
@@ -5391,6 +5394,31 @@ LEAN_COMMAND_NAMES = [
 ]
 
 
+@pytest.mark.parametrize(
+    "command_path",
+    [
+        CORE_CONSTITUTION_COMMAND,
+        LEAN_PRESET_DIR / "commands" / "speckit.constitution.md",
+    ],
+    ids=["core", "lean"],
+)
+def test_constitution_commands_guard_against_non_governance_work(command_path):
+    """Constitution commands defer non-governance work instead of executing it."""
+    content = command_path.read_text()
+    lower_content = content.lower()
+    normalized_content = " ".join(lower_content.split())
+
+    assert "## Scope Guard" in content
+    assert "**MUST NOT**" in content
+    assert "Classify every part" in content
+    assert "application source files" in content
+    assert "non-governance intent" in content
+    assert "`Next Actions`" in content
+    assert "__SPECKIT_COMMAND_SPECIFY__" in content
+    assert "omit" in lower_content
+    assert "do not invoke it" in normalized_content or "without invoking it" in normalized_content
+
+
 class TestLeanPreset:
     """Tests for the lean preset that ships with the repo."""
 
@@ -7458,10 +7486,10 @@ def test_preset_wrapper_resolves_ghes_asset_when_host_configured(tmp_path, monke
     def fake_open(url, timeout=None, extra_headers=None):
         captured.append(url)
         resp = MagicMock()
-        resp.read.return_value = json.dumps({
+        resp.read.side_effect = io.BytesIO(json.dumps({
             "assets": [{"name": "pack.zip",
                         "url": "https://ghes.example/api/v3/repos/o/r/releases/assets/9"}]
-        }).encode()
+        }).encode()).read
         yield resp
 
     monkeypatch.setattr(catalog, "_open_url", fake_open)

tests/test_self_upgrade_detection.py

@@ -13,6 +13,7 @@ import specify_cli
 from specify_cli import app
 
 from tests.self_upgrade_helpers import (
+    route_opener_open_through_urlopen,  # noqa: F401 (autouse fixture)
     _InstallMethod,
     _assemble_installer_argv,
     _completed_process,

tests/test_self_upgrade_execution.py

@@ -7,6 +7,7 @@ from unittest.mock import patch
 from specify_cli import app
 
 from tests.self_upgrade_helpers import (
+    route_opener_open_through_urlopen,  # noqa: F401 (autouse fixture)
     _completed_process,
     mock_urlopen_response,
     requires_posix,

tests/test_self_upgrade_guidance.py

@@ -6,6 +6,7 @@ from specify_cli import app
 
 from tests.self_upgrade_helpers import (
     mock_urlopen_response,
+    route_opener_open_through_urlopen,  # noqa: F401 (autouse fixture)
     runner,
     strip_ansi,
 )

tests/test_self_upgrade_verification.py

@@ -8,6 +8,7 @@ import specify_cli
 from specify_cli import app
 
 from tests.self_upgrade_helpers import (
+    route_opener_open_through_urlopen,  # noqa: F401 (autouse fixture)
     SENTINEL_GH_TOKEN,
     SENTINEL_GITHUB_TOKEN,
     _InstallMethod,

tests/test_specify_template_numbering.py

@@ -0,0 +1,39 @@
+"""Regression tests for top-level step numbering in specify.md."""
+
+import re
+from pathlib import Path
+
+REPO_ROOT = Path(__file__).parent.parent
+SPECIFY_TEMPLATE = REPO_ROOT / "templates" / "commands" / "specify.md"
+MAIN_LIST_START = "Given that feature description, do this:"
+MAIN_LIST_END = "## Mandatory Post-Execution Hooks"
+
+
+def _main_execution_ordinals(text: str) -> list[int]:
+    """Extract top-level ordinals from the main execution flow."""
+    _, start, execution_flow = text.partition(MAIN_LIST_START)
+    execution_flow, end, _ = execution_flow.partition(MAIN_LIST_END)
+    if not start or not end:
+        return []
+
+    return [
+        int(match.group(1))
+        for line in execution_flow.splitlines()
+        if (match := re.match(r"^(\d+)\. ", line))
+    ]
+
+
+def test_main_execution_list_has_no_duplicate_ordinals():
+    """The main execution list must not reuse a step number."""
+    ordinals = _main_execution_ordinals(SPECIFY_TEMPLATE.read_text(encoding="utf-8"))
+    duplicates = {ordinal for ordinal in ordinals if ordinals.count(ordinal) > 1}
+
+    assert not duplicates, f"Duplicate top-level ordinals found: {sorted(duplicates)}"
+
+
+def test_main_execution_list_is_sequential():
+    """The main execution list must run from 1 through N without gaps."""
+    ordinals = _main_execution_ordinals(SPECIFY_TEMPLATE.read_text(encoding="utf-8"))
+
+    assert ordinals, "Could not find the main execution list in specify.md"
+    assert ordinals == list(range(1, 9))

tests/test_upgrade.py

@@ -2,11 +2,12 @@
 
 Network isolation contract (SC-004 / FR-014): every test that exercises
 `specify self check` or `_fetch_latest_release_tag()` MUST mock the outbound
-urllib path it expects (`urlopen` for unauthenticated requests, `build_opener`
-for authenticated requests) so no real outbound call ever reaches api.github.com.
-Tests for non-network `self upgrade` behavior should keep that contract explicit
-with local mocks. Run this module under `pytest-socket` (if installed) with
-`--disable-socket` as an extra safety net.
+urllib path so no real call reaches api.github.com. Production always uses an
+isolated `build_opener`; this module's autouse fixture routes its `open()` back
+through the locally mocked `urlopen`. Tests for non-network `self upgrade`
+behavior should keep that contract explicit with local mocks. Run this module
+under `pytest-socket` (if installed) with `--disable-socket` as an extra safety
+net.
 """
 
 import urllib.error
@@ -17,6 +18,7 @@ import pytest
 from typer.testing import CliRunner
 
 from specify_cli import app
+from specify_cli._download_security import read_response_limited as _real_read_response_limited
 from specify_cli._version import (
     _fetch_latest_release_tag,
     _get_installed_version,
@@ -24,7 +26,10 @@ from specify_cli._version import (
     _normalize_tag,
 )
 from tests.conftest import strip_ansi
-from tests.http_helpers import mock_urlopen_response
+from tests.http_helpers import (
+    mock_urlopen_response,
+    route_opener_open_through_urlopen,  # noqa: F401 (autouse fixture)
+)
 
 runner = CliRunner()
 
@@ -235,6 +240,46 @@ class TestFailureCategorization:
                 _fetch_latest_release_tag()
 
 
+class TestBoundedRead:
+    """Regression test for the read_response_limited hardening.
+
+    A future refactor could silently revert `_fetch_latest_release_tag` to
+    `resp.read()` (the unbounded form) — this test pins the contract that
+    the response body is read through ``read_response_limited`` with a
+    bounded ``max_bytes``.
+    """
+
+    def test_response_body_is_bounded(self):
+        recorded: dict[str, int | str] = {}
+
+        def _spy(response, *, max_bytes: int, label: str, **kwargs):
+            # max_bytes and label are keyword-only with no defaults: if the
+            # caller forgets to pass either, the call raises TypeError here
+            # (instead of recording a misleading None).
+            recorded["max_bytes"] = max_bytes
+            recorded["label"] = label
+            # Forward to the real implementation so the function under test
+            # still gets a parseable body.
+            return _real_read_response_limited(
+                response, max_bytes=max_bytes, label=label, **kwargs
+            )
+
+        with patch(
+            "specify_cli.authentication.http.urllib.request.urlopen",
+            return_value=mock_urlopen_response({"tag_name": "v9.9.9"}),
+        ), patch("specify_cli._version.read_response_limited", side_effect=_spy):
+            tag, reason = _fetch_latest_release_tag()
+
+        assert tag == "v9.9.9"
+        assert reason is None
+        # The cap (1 MiB) is a deliberate ceiling for the GitHub release
+        # JSON — keep it explicit so a future refactor that drops the
+        # `max_bytes=` argument fails this test instead of regressing
+        # silently to the default.
+        assert recorded["max_bytes"] == 1024 * 1024
+        assert recorded["label"] == "GitHub latest release"
+
+
 _FAILURE_CASES = [
     ("offline or timeout", urllib.error.URLError("down")),
     (_RATE_LIMITED_REASON, _http_error(403)),

tests/test_workflows.py

@@ -404,6 +404,17 @@ class TestExpressions:
         assert evaluate_expression('{{ [["a", "b"], "c"] }}', ctx) == [["a", "b"], "c"]
         assert evaluate_expression("{{ [[1, 2], [3, 4]] }}", ctx) == [[1, 2], [3, 4]]
 
+    def test_list_literal_ignores_trailing_and_empty_commas(self):
+        from specify_cli.workflows.expressions import evaluate_expression
+        from specify_cli.workflows.base import StepContext
+
+        ctx = StepContext()
+        # A trailing comma must not append a spurious None element.
+        assert evaluate_expression("{{ [1, 2,] }}", ctx) == [1, 2]
+        assert evaluate_expression("{{ [1,, 2] }}", ctx) == [1, 2]
+        # …but an intentional empty-string element is still preserved.
+        assert evaluate_expression("{{ ['', 'a'] }}", ctx) == ["", "a"]
+
     def test_operator_splitting_is_quote_aware(self):
         from specify_cli.workflows.expressions import (
             evaluate_condition,
@@ -1162,6 +1173,21 @@ class TestCommandStep:
             result = step.execute(config, ctx)
         assert result.output["integration"] == "gemini"
 
+    def test_execute_non_string_integration_fails_cleanly(self):
+        """A non-string integration (e.g. a list from an expression that resolved
+        to one) must FAIL the step cleanly, not crash the run with
+        'TypeError: unhashable type: list' from get_integration's dict lookup."""
+        from specify_cli.workflows.steps.command import CommandStep
+        from specify_cli.workflows.base import StepContext, StepStatus
+
+        step = CommandStep()
+        config = {
+            "id": "s", "command": "speckit.plan",
+            "integration": ["claude"], "input": {},
+        }
+        result = step.execute(config, StepContext())
+        assert result.status == StepStatus.FAILED
+
     def test_step_override_model(self):
         from unittest.mock import patch
         from specify_cli.workflows.steps.command import CommandStep
@@ -1359,6 +1385,20 @@ class TestPromptStep:
         assert result.output["integration"] == "claude"
         assert result.output["dispatched"] is False
 
+    def test_execute_non_string_integration_fails_cleanly(self):
+        """A non-string integration must FAIL the step cleanly, not crash with
+        'TypeError: unhashable type: list' from get_integration's dict lookup."""
+        from specify_cli.workflows.steps.prompt import PromptStep
+        from specify_cli.workflows.base import StepContext, StepStatus
+
+        step = PromptStep()
+        config = {
+            "id": "p", "type": "prompt", "prompt": "do it",
+            "integration": ["claude"],
+        }
+        result = step.execute(config, StepContext())
+        assert result.status == StepStatus.FAILED
+
     def test_execute_with_step_integration(self):
         from unittest.mock import patch
         from specify_cli.workflows.steps.prompt import PromptStep
@@ -1895,6 +1935,14 @@ def _force_gate_stdin(monkeypatch, *, tty: bool):
 class TestInitStep:
     """Test the init step type."""
 
+    def test_docstring_lists_every_valid_script_type(self):
+        # The `script` field docstring must not contradict the step's own
+        # VALID_SCRIPT_TYPES (which includes 'py'); validate() accepts all three.
+        from specify_cli.workflows.steps.init import InitStep, VALID_SCRIPT_TYPES
+
+        for script_type in VALID_SCRIPT_TYPES:
+            assert f"``{script_type}``" in InitStep.__doc__
+
     def test_builds_here_argv_and_bootstraps(self, tmp_path):
         from specify_cli.workflows.steps.init import InitStep
         from specify_cli.workflows.base import StepContext, StepStatus
@@ -2060,6 +2108,15 @@ class TestInitStep:
 class TestGateStep:
     """Test the gate step type."""
 
+    def test_docstring_lists_every_on_reject_behaviour(self):
+        # The docstring must not contradict validate()/execute(): on_reject
+        # accepts 'abort', 'skip', AND 'retry' (execute() has a dedicated
+        # retry -> PAUSED branch), but the summary omitted 'retry'.
+        from specify_cli.workflows.steps.gate import GateStep
+
+        for behaviour in ("abort", "skip", "retry"):
+            assert behaviour in GateStep.__doc__
+
     @pytest.fixture(autouse=True)
     def _non_tty_stdin_by_default(self, monkeypatch):
         # Default every gate test to a non-TTY stdin so none can drop into
@@ -2145,6 +2202,19 @@ class TestGateStep:
         assert result.status == StepStatus.COMPLETED
         assert result.output["choice"] == "approve"
 
+    def test_interactive_prompt_rejects_non_decimal_digit(self, monkeypatch, capsys):
+        """A Unicode digit int() can't parse — e.g. the superscript '²', which
+        str.isdigit() accepts but int() rejects — must be treated as an invalid
+        choice, not crash the prompt loop with an uncaught ValueError."""
+        from specify_cli.workflows.steps.gate import GateStep
+
+        _force_gate_stdin(monkeypatch, tty=True)
+        inputs = iter(["²", "1"])  # superscript-two, then a real "1"
+        monkeypatch.setattr("builtins.input", lambda _prompt="": next(inputs))
+
+        choice = GateStep._prompt("Review the spec.", ["approve", "reject"])
+        assert choice == "approve"
+
     def test_interactive_prompt_missing_show_file_does_not_crash(
         self, tmp_path, monkeypatch, capsys
     ):
@@ -7733,6 +7803,63 @@ class TestWorkflowRemoveGuard:
         assert "[stage]permissiondenied" in output_compact
         assert "[reg]diskfull" in output_compact
 
+
+class TestWorkflowAddCaseInsensitiveSuffix:
+    """`workflow add` must detect a local YAML file case-insensitively, matching
+    `workflow run` (_commands.py:workflow_run) and the engine loader
+    (engine.py:WorkflowEngine.load_workflow), which both use `.suffix.lower()`.
+    Without it, `workflow run Sample.YAML` works but `workflow add Sample.YAML`
+    fails — an add/run inconsistency for an uppercase extension."""
+
+    def test_plain_path_accepts_uppercase_extension(self, temp_dir, monkeypatch, sample_workflow_yaml):
+        from typer.testing import CliRunner
+        from specify_cli import app
+
+        (temp_dir / ".specify" / "workflows").mkdir(parents=True)
+        src = temp_dir / "Sample.YAML"
+        src.write_text(sample_workflow_yaml, encoding="utf-8")
+
+        monkeypatch.chdir(temp_dir)
+        result = CliRunner().invoke(app, ["workflow", "add", str(src)])
+
+        # Before the fix: `.suffix in (...)` is case-sensitive, so ".YAML" is not
+        # recognized as a local file; the path falls through to catalog lookup
+        # and fails. After the fix it installs like the lowercase happy path.
+        assert result.exit_code == 0, result.output
+        assert "installed" in result.output
+
+    def test_dev_path_accepts_uppercase_extension(self, temp_dir, monkeypatch, sample_workflow_yaml):
+        from typer.testing import CliRunner
+        from specify_cli import app
+
+        (temp_dir / ".specify" / "workflows").mkdir(parents=True)
+        src = temp_dir / "Sample.YAML"
+        src.write_text(sample_workflow_yaml, encoding="utf-8")
+
+        monkeypatch.chdir(temp_dir)
+        result = CliRunner().invoke(app, ["workflow", "add", "--dev", str(src)])
+
+        # Before the fix the --dev branch rejects ".YAML" with
+        # "--dev source must be a workflow YAML file ...".
+        assert result.exit_code == 0, result.output
+        assert "installed" in result.output
+
+    def test_lowercase_extension_still_installs(self, temp_dir, monkeypatch, sample_workflow_yaml):
+        """Happy path (lowercase .yml) is unchanged by the case-normalization."""
+        from typer.testing import CliRunner
+        from specify_cli import app
+
+        (temp_dir / ".specify" / "workflows").mkdir(parents=True)
+        src = temp_dir / "sample.yml"
+        src.write_text(sample_workflow_yaml, encoding="utf-8")
+
+        monkeypatch.chdir(temp_dir)
+        result = CliRunner().invoke(app, ["workflow", "add", str(src)])
+
+        assert result.exit_code == 0, result.output
+        assert "installed" in result.output
+
+
 class TestWorkflowAddSymlinkGuard:
     def test_add_malformed_ipv6_url_exits_cleanly(self, temp_dir, monkeypatch):
         """A malformed IPv6 URL must produce a clean error, not a ValueError traceback."""
@@ -8548,18 +8675,15 @@ steps:
         class FakeResponse:
             def __init__(self, data, url=None):
                 self._data = data
+                self._pos = 0
                 self._url = url or "https://api.github.com/repos/org/repo/releases/assets/42"
 
-            def read(self, amt=None):
-                if not hasattr(self, "_pos"):
-                    self._pos = 0
-                if amt is None:
-                    chunk = self._data[self._pos :]
-                    self._pos = len(self._data)
-                    return chunk
-                chunk = self._data[self._pos : self._pos + amt]
-                self._pos += len(chunk)
-                return chunk
+            def read(self, size=-1):
+                if size < 0:
+                    size = len(self._data) - self._pos
+                out = self._data[self._pos : self._pos + size]
+                self._pos += len(out)
+                return out
 
             def geturl(self):
                 return self._url
@@ -8619,18 +8743,15 @@ steps:
         class FakeResponse:
             def __init__(self, data, url=None):
                 self._data = data
+                self._pos = 0
                 self._url = url or "https://api.github.com/repos/org/repo/releases/assets/42"
 
-            def read(self, amt=None):
-                if not hasattr(self, "_pos"):
-                    self._pos = 0
-                if amt is None:
-                    chunk = self._data[self._pos :]
-                    self._pos = len(self._data)
-                    return chunk
-                chunk = self._data[self._pos : self._pos + amt]
-                self._pos += len(chunk)
-                return chunk
+            def read(self, size=-1):
+                if size < 0:
+                    size = len(self._data) - self._pos
+                out = self._data[self._pos : self._pos + size]
+                self._pos += len(out)
+                return out
 
             def geturl(self):
                 return self._url
@@ -8670,18 +8791,15 @@ steps:
         class FakeResponse:
             def __init__(self, data, url=None):
                 self._data = data
+                self._pos = 0
                 self._url = url or "https://api.github.com/repos/org/repo/releases/assets/55"
 
-            def read(self, amt=None):
-                if not hasattr(self, "_pos"):
-                    self._pos = 0
-                if amt is None:
-                    chunk = self._data[self._pos :]
-                    self._pos = len(self._data)
-                    return chunk
-                chunk = self._data[self._pos : self._pos + amt]
-                self._pos += len(chunk)
-                return chunk
+            def read(self, size=-1):
+                if size < 0:
+                    size = len(self._data) - self._pos
+                out = self._data[self._pos : self._pos + size]
+                self._pos += len(out)
+                return out
 
             def geturl(self):
                 return self._url
@@ -8763,18 +8881,15 @@ steps:
         class FakeResponse:
             def __init__(self, data, url=None):
                 self._data = data
+                self._pos = 0
                 self._url = url or "https://ghes.example/api/v3/repos/org/repo/releases/assets/42"
 
-            def read(self, amt=None):
-                if not hasattr(self, "_pos"):
-                    self._pos = 0
-                if amt is None:
-                    chunk = self._data[self._pos :]
-                    self._pos = len(self._data)
-                    return chunk
-                chunk = self._data[self._pos : self._pos + amt]
-                self._pos += len(chunk)
-                return chunk
+            def read(self, size=-1):
+                if size < 0:
+                    size = len(self._data) - self._pos
+                out = self._data[self._pos : self._pos + size]
+                self._pos += len(out)
+                return out
 
             def geturl(self):
                 return self._url
@@ -8826,18 +8941,15 @@ steps:
         class FakeResponse:
             def __init__(self, data, url=None):
                 self._data = data
+                self._pos = 0
                 self._url = url or "https://ghes.example/api/v3/repos/org/repo/releases/assets/55"
 
-            def read(self, amt=None):
-                if not hasattr(self, "_pos"):
-                    self._pos = 0
-                if amt is None:
-                    chunk = self._data[self._pos :]
-                    self._pos = len(self._data)
-                    return chunk
-                chunk = self._data[self._pos : self._pos + amt]
-                self._pos += len(chunk)
-                return chunk
+            def read(self, size=-1):
+                if size < 0:
+                    size = len(self._data) - self._pos
+                out = self._data[self._pos : self._pos + size]
+                self._pos += len(out)
+                return out
 
             def geturl(self):
                 return self._url
@@ -9973,6 +10085,17 @@ steps:
         registry.add("align-wf", {"version": "1.0.0", "source": "catalog"})
         assert registry.get("align-wf")["version"] == "1.0.0"
 
+    def test_step_registry_add_survives_non_dict_existing_entry(self, project_dir):
+        """StepRegistry.add must treat a corrupted non-dict existing entry as
+        absent rather than crash on existing.get() (parity with
+        WorkflowRegistry.add)."""
+        from specify_cli.workflows.catalog import StepRegistry
+
+        registry = StepRegistry(project_dir)
+        registry.data["steps"]["my-step"] = "corrupted"
+        registry.add("my-step", {"version": "1.0.0"})
+        assert registry.get("my-step")["version"] == "1.0.0"
+
     @pytest.mark.parametrize(
         "contents",
         [
@@ -11630,6 +11753,10 @@ steps:
             _reject_insecure_download_redirect(
                 "https://example.com/wf.yml", "http://localhost:8000/wf.yml"
             )
+        with pytest.raises(urllib.error.URLError):
+            _reject_insecure_download_redirect(
+                "https://example.com/wf.yml", "https://127.0.0.2/wf.yml"
+            )
         # Allowed: HTTPS anywhere, or loopback HTTP that stays on loopback HTTP.
         _reject_insecure_download_redirect(
             "https://example.com/wf.yml", "https://cdn.example.com/wf.yml"
@@ -11640,6 +11767,9 @@ steps:
         _reject_insecure_download_redirect(
             "http://127.0.0.1/source.yml", "http://127.0.0.1/wf.yml"
         )
+        _reject_insecure_download_redirect(
+            "http://127.0.0.2/source.yml", "http://127.255.255.254/wf.yml"
+        )
 
     def test_add_from_url_passes_redirect_validator(self, project_dir, monkeypatch):
         from unittest.mock import patch

tests/unit/test_bundler_catalog_config.py

@@ -253,6 +253,47 @@ def test_add_source_wraps_invalid_ipv6_as_bundler_error(tmp_path: Path):
         cc.add_source(project, "https://[::1/c.json", policy="install-allowed", priority=50)
 
 
+def test_add_source_wraps_bracketed_non_ip_host_as_bundler_error(tmp_path: Path):
+    # A bracketed-but-invalid IPv6 authority (e.g. "https://[not-an-ip]/c.json")
+    # parses cleanly under urlparse() on Python < 3.14 and only raises ValueError
+    # lazily on the first .hostname access; the raise moved eager into urlparse()
+    # in 3.14. add_source must surface its own BundlerError on every supported
+    # version, never leak a raw ValueError past the CLI's `except BundlerError`.
+    project = tmp_path / "proj"
+    (project / ".specify").mkdir(parents=True)
+    with pytest.raises(BundlerError, match="Invalid catalog url"):
+        cc.add_source(project, "https://[not-an-ip]/c.json", policy="install-allowed", priority=50)
+
+
+def test_add_source_wraps_lazy_hostname_valueerror(tmp_path: Path, monkeypatch):
+    # Simulate the Python < 3.14 shape explicitly (independent of the running
+    # interpreter): urlparse() succeeds but .hostname raises ValueError lazily.
+    # This is the exact path the fix guards; it fails with a raw ValueError if
+    # .hostname is read outside the try/except.
+    from urllib.parse import urlparse as _real_urlparse
+
+    class _LazyHostnameRaiser:
+        def __init__(self, parsed):
+            self._parsed = parsed
+
+        @property
+        def hostname(self):
+            raise ValueError("simulated lazy IPv6 hostname failure")
+
+        def __getattr__(self, name):
+            return getattr(self._parsed, name)
+
+    def _fake_urlparse(url, *args, **kwargs):
+        return _LazyHostnameRaiser(_real_urlparse(url, *args, **kwargs))
+
+    monkeypatch.setattr(cc, "urlparse", _fake_urlparse)
+
+    project = tmp_path / "proj"
+    (project / ".specify").mkdir(parents=True)
+    with pytest.raises(BundlerError, match="Invalid catalog url"):
+        cc.add_source(project, "https://example.com/c.json", policy="install-allowed", priority=50)
+
+
 def test_remove_source_does_not_crash_on_invalid_ipv6(tmp_path: Path):
     project = tmp_path / "proj"
     (project / ".specify").mkdir(parents=True)

tests/unit/test_bundler_packager.py

@@ -73,6 +73,23 @@ def test_build_is_deterministic(tmp_path: Path):
     assert first.artifact_path.read_bytes() == second.artifact_path.read_bytes()
 
 
+def test_member_order_is_platform_independent(tmp_path: Path):
+    # Members must be laid out in canonical POSIX-arcname order (the same key
+    # build_bundle uses to NAME them), not pathlib.Path order — which folds case
+    # on Windows and would otherwise reorder members across build hosts, breaking
+    # the byte-for-byte reproducibility guarantee. Mixed-case names make the
+    # difference observable: Path order on Windows groups differently than the
+    # canonical string sort.
+    bundle = _make_bundle(
+        tmp_path / "b",
+        extra_files={"Zeta.txt": "z", "apple.txt": "a", "Foo.txt": "f", "bar.txt": "b"},
+    )
+    result = build_bundle(bundle, output_dir=tmp_path / "out")
+    with zipfile.ZipFile(result.artifact_path) as archive:
+        names = archive.namelist()
+    assert names == sorted(names)
+
+
 def test_output_dir_inside_bundle_excludes_prior_artifacts(tmp_path: Path):
     bundle = _make_bundle(tmp_path / "b", extra_files={"a.txt": "a"})
     out_dir = bundle / "dist"

tests/unit/test_bundler_yamlio.py

@@ -0,0 +1,26 @@
+"""Unit tests for the bundler YAML I/O helpers."""
+from __future__ import annotations
+
+from pathlib import Path
+
+from specify_cli.bundler.lib.yamlio import dump_yaml, load_yaml
+
+
+def test_dump_yaml_preserves_unicode(tmp_path: Path):
+    # dump_yaml must write literal UTF-8, not \xNN / \uXXXX escapes, so bundle
+    # config stays human-readable — matching _utils.dump_frontmatter and the
+    # extensions/presets config writers (all allow_unicode=True).
+    path = tmp_path / "f.yml"
+    data = {"note": "café-münchen", "url": "https://例え.example"}
+    dump_yaml(path, data)
+    raw = path.read_text(encoding="utf-8")
+    assert "café-münchen" in raw
+    assert "例え" in raw
+    assert "\\x" not in raw and "\\u" not in raw
+
+
+def test_dump_yaml_round_trips_unicode(tmp_path: Path):
+    path = tmp_path / "f.yml"
+    data = {"note": "café", "city": "münchen"}
+    dump_yaml(path, data)
+    assert load_yaml(path) == data

tests/workflows/test_overlay_merge.py

@@ -162,6 +162,23 @@ class TestMergeSteps:
             ComposedStep("low-step", "project:low"),
         ]
 
+    def test_merge_steps_multiple_insert_after_same_overlay_preserves_order(self):
+        # Two insert_after edits from ONE overlay on the same anchor must keep
+        # their declared order (a, x, y, b) — mirroring insert_before. The old
+        # reversed(edits) over the flat list flipped them to (a, y, x, b).
+        base = [_step("a"), _step("b")]
+        overlay = Overlay(
+            id="ov1",
+            extends="wf",
+            priority=10,
+            edits=[
+                OverlayEdit("insert_after", "a", _step("x")),
+                OverlayEdit("insert_after", "a", _step("y")),
+            ],
+        )
+        steps, _ = merge_steps(base, [_layer(overlay, "project:ov1")])
+        assert [s["id"] for s in steps] == ["a", "x", "y", "b"]
+
     def test_merge_steps_replace_wins_over_insert(self):
         """Overlays apply to the original tree only; targeting an overlay-introduced step raises."""
         base = [_step("a")]