fix(cursor-agent): enable headless CLI dispatch end-to-end (-p --trust --approve-mcps --force + Windows .cmd shim resolution) (#2631)

* fix(cursor-agent): enable CLI dispatch via ``-p --trust`` headless mode

Restores the ability for ``specify workflow run`` to dispatch the
cursor-agent CLI, complementing the existing in-IDE skill flow.
Without this fix, ``specify workflow run speckit --input
integration=cursor-agent ...`` fails with a misleading
``CLI not found or not installed`` error even when the CLI is
installed (since cursor-agent had ``requires_cli=False`` and an
unset ``build_exec_args``).

The cursor-agent CLI (>= 2026.05.16) supports headless execution
via ``-p`` (print mode with full tool access including write/shell)
and ``--trust`` (bypass Workspace Trust prompt). Without ``--trust``
the CLI exits non-zero in non-TTY contexts (verified locally).

Changes to ``src/specify_cli/integrations/cursor_agent/__init__.py``:

* ``config.requires_cli``: ``False`` -> ``True``
* ``config.install_url``: ``None`` -> Cursor CLI docs URL
* Override ``build_exec_args()`` to emit
  ``[cursor-agent, -p, --trust, <prompt>, ...]``
  with optional ``--model`` and ``--output-format json`` flags,
  mirroring the shape used by ``claude``/``codex``/``gemini``.

Tests:

* 34 existing cursor-agent tests still pass.
* 6 new tests in ``TestCursorAgentCliDispatch`` pin
  ``requires_cli``, ``install_url``, and the exact argv shape
  (default, text-output, with-model, and the hyphenated skill
  invocation form ``/speckit-<name>``).
* Full repo: 1085 / 1085 passed, no regressions.

Fixes #2629

Co-authored-by: Cursor <cursoragent@cursor.com>

* fix(integrations): resolve ``.cmd``/``.bat`` shims before subprocess.run

On Windows, ``shutil.which`` honors ``PATHEXT`` and locates wrappers
like ``cursor-agent.cmd`` and ``codex.cmd``, but Python's
``subprocess.run`` calls ``CreateProcess`` which does **not** consult
``PATHEXT`` and therefore fails with ``WinError 2`` on a bare argv
like ``[cursor-agent, ...]``.

Resolve ``exec_args[0]`` via ``shutil.which`` in
``IntegrationBase.dispatch_command`` so ``.cmd``/``.bat`` shims work
transparently. On POSIX this is a no-op for absolute paths and a
harmless lookup otherwise.

Verified locally on Windows 10 + cursor-agent 2026.05.16:
without this fix, ``specify workflow run speckit --input
integration=cursor-agent`` fails with ``FileNotFoundError`` even
after the cursor-agent integration starts producing valid exec
args (per the prior commit on this branch).

Tests:

* New: 2 cursor-agent tests pin the shim-resolution + passthrough
  behavior (``test_dispatch_command_resolves_cmd_shim_for_subprocess``
  and ``test_dispatch_command_passthrough_when_shutil_which_finds_nothing``).
* Updated: ``tests/test_workflows.py::TestCommandStep::test_dispatch_with_mock_cli``
  was mocking ``shutil.which`` only at the ``command`` step level
  and not at the ``base`` level, which made it environment-sensitive
  (fails locally when the real ``claude`` CLI is on PATH).  Added the
  matching base-level patch and updated the argv-assertion to reflect
  the resolved path. ``test_dispatch_failure_returns_failed_status``
  gets the same patch for consistency.
* Full repo: 2867 passed, 0 regression from this PR. The 12 remaining
  pre-existing failures are unrelated Windows ``symlink`` privilege
  failures (``WinError 1314``) on a non-admin Windows runner.

Co-authored-by: Cursor <cursoragent@cursor.com>

* fix(cursor-agent): inject --approve-mcps --force for headless MCP/tool access

The previous commit (1c55988) wired up ``-p --trust`` so the CLI launches
in headless mode without the Workspace Trust prompt, but that alone is
not enough to let ``specify workflow run`` drive a real speckit feature
end-to-end with cursor-agent on Windows. Two more flags are required:

* ``--approve-mcps``: without it, every MCP server configured in
  ``.cursor/mcp.json`` stays ``not loaded (needs approval)``, and any
  tool call against them is silently dropped. We hit this immediately
  trying to read a DingTalk PRD from a remote MCP server during the
  ``/speckit-specify`` step.
* ``--force``: without it, the agent halts on the first tool-call
  approval prompt (the tool call gets rejected and the workflow exits
  non-zero with a misleading message). With ``--force`` cursor-agent
  matches the implicit "trusted environment" semantics that ``claude -p``
  and ``codex --exec`` already have by default -- which is the right
  semantics for an unattended ``specify workflow run`` invocation.

Verified end-to-end on Windows 10 + cursor-agent 2026.05.16-0338208:

* ``cursor-agent -p --trust --approve-mcps --force --output-format text``
  + a ``/speckit-specify`` prompt that included a DingTalk URL produced
  a full spec.md (31.5 KB) plus checklists/requirements.md in ~10.7 min,
  reading the source PRD through the ``dingtalk-doc`` remote MCP server,
  deciding the ``specs/`` subpath itself, and updating
  ``.specify/feature.json`` and ``specs/menu-dictionary.md`` along the
  way -- no human-in-the-loop, no source PRD ever touched the filesystem.
* Without ``--approve-mcps`` the same prompt errors with the tool call
  rejected message; without ``--force`` the agent stops at the first
  non-MCP tool call.

Tests:

* ``test_build_exec_args_*`` updated to pin the new four-flag prefix.
* New ``test_build_exec_args_contains_mandatory_headless_flags`` asserts
  the four flags are always present together.
* ``test_dispatch_command_resolves_cmd_shim_for_subprocess`` updated to
  match the new argv layout.
* All 43 cursor-agent tests pass; no other tests touched.

Co-authored-by: Cursor <cursoragent@cursor.com>

* refactor(cursor-agent): express dispatch support via build_exec_args() instead of requires_cli

Co-authored-by: Cursor <cursoragent@cursor.com>

* test(cursor-agent): use urlparse hostname check and cover dispatch without requires_cli

Co-authored-by: Cursor <cursoragent@cursor.com>

* Potential fix for pull request finding

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

---------

Co-authored-by: 刘一 <liuyi@oureman.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

src/specify_cli/integrations/base.py

@@ -270,6 +270,16 @@ class IntegrationBase(ABC):
             )
             raise NotImplementedError(msg)
 
+        # Windows: ``subprocess.run`` calls ``CreateProcess`` which does not
+        # consult ``PATHEXT``, so a bare command name like ``cursor-agent``
+        # that resolves to ``cursor-agent.cmd`` fails with ``WinError 2``.
+        # Resolve via ``shutil.which`` (which does honor ``PATHEXT``) so
+        # ``.cmd``/``.bat`` shims work transparently.  On POSIX this is a
+        # no-op for absolute paths and a harmless lookup otherwise.
+        resolved = shutil.which(exec_args[0])
+        if resolved:
+            exec_args = [resolved, *exec_args[1:]]
+
         cwd = str(project_root) if project_root else None
 
         if stream:

src/specify_cli/integrations/cursor_agent/__init__.py

@@ -2,6 +2,12 @@
 
 Cursor Agent uses the ``.cursor/skills/speckit-<name>/SKILL.md`` layout.
 Commands are deprecated; ``--skills`` defaults to ``True``.
+
+The IDE/skills flow is the primary path and works without the
+``cursor-agent`` CLI being installed (``requires_cli=False``).  Workflow
+dispatch via ``cursor-agent -p --trust --approve-mcps --force <prompt>``
+is offered as an opt-in capability — the presence of ``build_exec_args()``
+is what indicates dispatch support, mirroring ``CopilotIntegration``.
 """
 
 from __future__ import annotations
@@ -15,7 +21,12 @@ class CursorAgentIntegration(SkillsIntegration):
         "name": "Cursor",
         "folder": ".cursor/",
         "commands_subdir": "skills",
-        "install_url": None,
+        "install_url": "https://docs.cursor.com/en/cli/overview",
+        # IDE-first integration: ``specify init --ai cursor-agent`` must
+        # work without the ``cursor-agent`` CLI installed (the IDE flow
+        # uses skills directly).  Workflow dispatch additionally requires
+        # the CLI on PATH, but that's enforced at dispatch time via
+        # ``shutil.which`` rather than as a hard ``specify init`` precheck.
         "requires_cli": False,
     }
     registrar_config = {
@@ -28,6 +39,50 @@ class CursorAgentIntegration(SkillsIntegration):
     context_file = ".cursor/rules/specify-rules.mdc"
     multi_install_safe = True
 
+    def build_exec_args(
+        self,
+        prompt: str,
+        *,
+        model: str | None = None,
+        output_json: bool = True,
+    ) -> list[str] | None:
+        """Build CLI arguments for non-interactive ``cursor-agent`` execution.
+
+        Always returns argv (no ``requires_cli`` guard) so workflow
+        dispatch is supported even though the integration's ``config``
+        sets ``requires_cli=False`` to keep the IDE-only flow unblocked.
+        This mirrors ``CopilotIntegration``: dispatch support is signalled
+        by overriding ``build_exec_args()``, not by the ``requires_cli``
+        flag (which is reserved for the ``specify init`` precheck).
+
+        Mandatory headless flags:
+
+        * ``-p`` — print/headless mode (access to all tools)
+        * ``--trust`` — bypass Workspace Trust prompt (CLI exits non-zero
+          otherwise)
+        * ``--approve-mcps`` — auto-approve MCP server loading (otherwise
+          MCP servers stay ``not loaded (needs approval)`` and tool calls
+          to them are silently dropped)
+        * ``--force`` — auto-approve tool invocations (shell/write/MCP),
+          matching the implicit "trusted environment" semantics that other
+          integrations (``claude -p``, ``codex --exec``) get by default
+
+        Together these are the minimum set required to make
+        ``specify workflow run speckit --input integration=cursor-agent``
+        behave the same way as it does for ``claude`` / ``codex``.
+        Verified locally: with ``--approve-mcps --force`` the agent can
+        call any configured MCP server (e.g. ``dingtalk-doc``) and write
+        files during ``/speckit-*`` skill execution; without them the run
+        either drops tool calls or exits non-zero on the first approval
+        prompt.
+        """
+        args = [self.key, "-p", "--trust", "--approve-mcps", "--force", prompt]
+        if model:
+            args.extend(["--model", model])
+        if output_json:
+            args.extend(["--output-format", "json"])
+        return args
+
     @classmethod
     def options(cls) -> list[IntegrationOption]:
         return [

tests/integrations/test_integration_cursor_agent.py

@@ -1,6 +1,7 @@
 """Tests for CursorAgentIntegration."""
 
 from pathlib import Path
+from urllib.parse import urlparse
 
 from specify_cli.integrations import get_integration
 from specify_cli.integrations.manifest import IntegrationManifest
@@ -106,3 +107,157 @@ class TestCursorAgentAutoPromote:
         assert result.exit_code == 0, f"init --ai cursor-agent failed: {result.output}"
         assert (target / ".cursor" / "skills" / "speckit-plan" / "SKILL.md").exists()
 
+
+class TestCursorAgentCliDispatch:
+    """Verify the CLI dispatch path for cursor-agent (issue #2629).
+
+    The ``cursor-agent`` CLI supports headless execution via ``-p`` (with
+    full tool access including write/shell) and requires ``--trust`` to
+    bypass the Workspace Trust prompt.  These tests pin the exact argv
+    shape that the workflow runner will use.
+    """
+
+    def test_requires_cli_is_false_for_ide_first_flow(self):
+        """``requires_cli`` must stay False so the IDE-only flow keeps working.
+
+        ``specify init --ai cursor-agent`` (without ``--ignore-agent-tools``)
+        treats ``requires_cli=True`` as a hard precheck and fails when the
+        ``cursor-agent`` CLI isn't on PATH — even though the Cursor IDE
+        / skills flow can run without it.  Workflow dispatch support is
+        signalled by overriding ``build_exec_args()`` instead, mirroring
+        ``CopilotIntegration``.
+        """
+        i = get_integration("cursor-agent")
+        assert i.config.get("requires_cli") is False
+
+    def test_install_url_is_set(self):
+        i = get_integration("cursor-agent")
+        url = i.config.get("install_url")
+        assert url is not None
+        # CodeQL: use a hostname comparison instead of a substring check
+        # to avoid the "Incomplete URL substring sanitization" warning
+        # (substring "cursor.com" can also appear in attacker-controlled
+        # positions of an arbitrary URL).
+        host = (urlparse(url).hostname or "").lower()
+        assert host == "cursor.com" or host.endswith(".cursor.com")
+
+    def test_build_exec_args_default_includes_headless_flags_and_json(self):
+        """Default argv emits the full headless flag set: -p --trust
+        --approve-mcps --force, then prompt, then --output-format json.
+        """
+        i = get_integration("cursor-agent")
+        args = i.build_exec_args("/speckit-specify some-feature")
+        assert args == [
+            "cursor-agent", "-p", "--trust", "--approve-mcps", "--force",
+            "/speckit-specify some-feature",
+            "--output-format", "json",
+        ]
+
+    def test_build_exec_args_text_output_omits_format(self):
+        i = get_integration("cursor-agent")
+        args = i.build_exec_args("/speckit-plan", output_json=False)
+        assert args == [
+            "cursor-agent", "-p", "--trust", "--approve-mcps", "--force",
+            "/speckit-plan",
+        ]
+
+    def test_build_exec_args_with_model(self):
+        i = get_integration("cursor-agent")
+        args = i.build_exec_args(
+            "/speckit-specify", model="sonnet-4-thinking", output_json=False
+        )
+        assert args == [
+            "cursor-agent", "-p", "--trust", "--approve-mcps", "--force",
+            "/speckit-specify",
+            "--model", "sonnet-4-thinking",
+        ]
+
+    def test_build_exec_args_contains_mandatory_headless_flags(self):
+        """The four headless flags must always appear together.
+
+        ``--approve-mcps`` is required so MCP servers (e.g. dingtalk-doc)
+        actually load in headless mode; ``--force`` is required so the
+        agent doesn't block on tool-call approval prompts during the
+        speckit workflow.  Together with ``-p`` and ``--trust`` they
+        bring cursor-agent's headless behaviour in line with
+        ``claude -p`` / ``codex --exec`` from spec-kit's perspective.
+        """
+        i = get_integration("cursor-agent")
+        args = i.build_exec_args("/speckit-implement", output_json=False)
+        for flag in ("-p", "--trust", "--approve-mcps", "--force"):
+            assert flag in args, f"missing mandatory headless flag: {flag}"
+
+    def test_build_exec_args_supports_dispatch_without_requires_cli(self):
+        """``build_exec_args`` must return argv even though ``requires_cli``
+        is ``False``.
+
+        ``CursorAgentIntegration`` opts out of the ``requires_cli`` hard
+        precheck (so ``specify init`` doesn't fail when the CLI isn't on
+        PATH) but still supports workflow dispatch.  The presence of a
+        non-``None`` argv from ``build_exec_args()`` is what the engine
+        keys off — pin that invariant.
+        """
+        i = get_integration("cursor-agent")
+        assert i.config.get("requires_cli") is False
+        argv = i.build_exec_args("/speckit-plan", output_json=False)
+        assert argv is not None
+        assert argv[0] == "cursor-agent"
+
+    def test_build_command_invocation_uses_hyphenated_skill_name(self):
+        """SkillsIntegration: /speckit-plan (not /speckit.plan)."""
+        i = get_integration("cursor-agent")
+        assert i.build_command_invocation("speckit.plan", "feature-x") == "/speckit-plan feature-x"
+        assert i.build_command_invocation("plan") == "/speckit-plan"
+
+    def test_dispatch_command_resolves_cmd_shim_for_subprocess(self):
+        """``.cmd`` shims must be resolved to their full path before ``subprocess.run``.
+
+        ``cursor-agent`` (and other npm-installed CLIs on Windows) ship as
+        ``cursor-agent.cmd`` wrappers.  ``shutil.which`` honors ``PATHEXT``
+        and finds them, but Python's ``subprocess.run`` calls
+        ``CreateProcess`` which does **not** consult ``PATHEXT`` and fails
+        with ``WinError 2`` on a bare ``["cursor-agent", ...]`` argv.  The
+        fix in ``base.py::dispatch_command`` resolves ``exec_args[0]`` via
+        ``shutil.which`` so the full ``.cmd`` path is what reaches
+        ``CreateProcess``.
+        """
+        from unittest.mock import patch, MagicMock
+        i = get_integration("cursor-agent")
+
+        mock_result = MagicMock()
+        mock_result.returncode = 0
+        mock_result.stdout = "ok"
+        mock_result.stderr = ""
+
+        fake_path = r"C:\Users\foo\AppData\Local\cursor-agent\cursor-agent.CMD"
+        with patch(
+            "specify_cli.integrations.base.shutil.which", return_value=fake_path
+        ), patch("subprocess.run", return_value=mock_result) as mock_run:
+            result = i.dispatch_command(
+                "speckit.plan", args="feature-x", stream=False, timeout=5
+            )
+
+        assert result["exit_code"] == 0
+        argv = mock_run.call_args[0][0]
+        assert argv[0] == fake_path, f"expected resolved .CMD path, got: {argv[0]!r}"
+        assert argv[1:6] == ["-p", "--trust", "--approve-mcps", "--force", "/speckit-plan feature-x"]
+
+    def test_dispatch_command_passthrough_when_shutil_which_finds_nothing(self):
+        """If ``shutil.which`` returns ``None``, leave argv unchanged so the
+        existing ``FileNotFoundError`` path remains observable to callers."""
+        from unittest.mock import patch, MagicMock
+        i = get_integration("cursor-agent")
+
+        mock_result = MagicMock()
+        mock_result.returncode = 0
+        mock_result.stdout = ""
+        mock_result.stderr = ""
+
+        with patch(
+            "specify_cli.integrations.base.shutil.which", return_value=None
+        ), patch("subprocess.run", return_value=mock_result) as mock_run:
+            i.dispatch_command("speckit.plan", stream=False, timeout=5)
+
+        argv = mock_run.call_args[0][0]
+        assert argv[0] == "cursor-agent"
+

tests/test_workflows.py

@@ -601,15 +601,18 @@ class TestCommandStep:
         mock_result.stderr = ""
 
         with patch("specify_cli.workflows.steps.command.shutil.which", return_value="/usr/local/bin/claude"), \
+             patch("specify_cli.integrations.base.shutil.which", return_value="/usr/local/bin/claude"), \
              patch("subprocess.run", return_value=mock_result) as mock_run:
             result = step.execute(config, ctx)
 
         assert result.status == StepStatus.COMPLETED
         assert result.output["dispatched"] is True
         assert result.output["exit_code"] == 0
-        # Verify the CLI was called with -p and the skill invocation
+        # Verify the CLI was called with the resolved path (via shutil.which,
+        # which honors PATHEXT for ``.cmd``/``.bat`` shims on Windows), then
+        # ``-p`` and the skill invocation.
         call_args = mock_run.call_args
-        assert call_args[0][0][0] == "claude"
+        assert call_args[0][0][0] == "/usr/local/bin/claude"
         assert call_args[0][0][1] == "-p"
         # Claude is a SkillsIntegration so uses /speckit-specify
         assert "/speckit-specify login" in call_args[0][0][2]
@@ -638,6 +641,7 @@ class TestCommandStep:
         mock_result.stderr = "API error"
 
         with patch("specify_cli.workflows.steps.command.shutil.which", return_value="/usr/local/bin/claude"), \
+             patch("specify_cli.integrations.base.shutil.which", return_value="/usr/local/bin/claude"), \
              patch("subprocess.run", return_value=mock_result):
             result = step.execute(config, ctx)