Fixesanthropics/claude-plugins-official#2048 — teams using Graphite
for stacked PRs (`gt create` / `gt modify` / `gt submit`) never get
the commit/push agentic review because the hook matcher only catches
literal `git commit` / `git push` Bash calls. gt shells out to git
as a subprocess, but the hook fires on Claude's top-level tool call,
which is `gt create` — not the `git commit` invocation inside the
gt subprocess that Claude Code never observes.
Per-edit pattern checks and end-of-turn Stop review still fire (those
don't depend on detecting the commit command), so the silent-coverage-
gap is bounded to the deepest review layer for Graphite users. Still:
that's exactly the layer designed to catch IDOR / auth-bypass /
cross-file SSRF, so the gap matters.
Semantic mapping (per the reporter):
gt create -> commit (like `git commit`)
gt modify -> commit + amend (like `git commit --amend`)
gt submit -> push (like `git push`)
Changes:
1. hooks/hooks.json: extend two PostToolUse `if` matchers.
"Bash(git commit:*)"
-> "Bash(git commit:*)|Bash(gt create:*)|Bash(gt modify:*)"
"Bash(git push:*)"
-> "Bash(git push:*)|Bash(gt submit:*)"
Without this, the hook subprocess never spawns for gt invocations
and the Python regex changes below are dead code.
2. hooks/security_reminder_hook.py: extend three regexes that classify
the bash command line.
_GIT_COMMIT_RE: now also matches `gt create` and `gt modify`.
Used at 4 sites (handler gate, multi-commit count, prompt
detection, event classification). Compound commands like
`gt create -am a && gt submit` now correctly trigger both the
commit and push paths.
_GIT_AMEND_RE: now also matches `gt modify` (semantically an
amend). The amend code path uses reflog to find the pre-amend
SHA and diff against THAT instead of HEAD~1 — same code path
now applies to `gt modify`.
_GIT_PUSH_RE: now also matches `gt submit`. Tolerates the same
`git -C path` / `git -c k=v` global options as before for the
git form; gt has its own flag layer that doesn't conflict.
Verified locally on macOS Python 3.13:
- JSON valid (hooks.json roundtrips).
- Existing 45 smoke + extensibility tests still pass.
- 76 new tests in test_gt_graphite_workflow.py (added to internal
test suite this PR doesn't ship — kept in sg-staging tests/ until
we have a story for shipping plugin tests publicly):
* 16 parametrized commit-match: native git commit variants +
all gt create / gt modify variants from the reporter's repro.
* 11 parametrized commit-reject: gt submit, gt log, gtoolkit
(word-boundary), agt create, etc.
* 9 parametrized amend-match: git commit --amend variants +
gt modify variants + chained git+gt.
* 7 parametrized amend-reject: regular git commit, gt create,
gt submit, echo'd substring noise.
* 11 parametrized push-match: git push variants + gt submit
variants + chained.
* 12 parametrized push-reject: git commit, gt log, gt fetch,
gt down, gt restack, gh pr create, agt submit.
* 3 compound-command class tests: git+gt mixtures trigger both
paths; gt modify chained with gt submit triggers
amend + push.
* 3 commit-invocation-count tests: gt commands contribute to
the multi-commit-detection findall count.
* 2 hooks.json static config tests: read the JSON, verify the
commit and push `if` clauses include the gt cases. Catches
the easy regression where someone updates the Python regex
but forgets to widen the matcher.
- 121/121 pass total (45 existing + 76 new) in 2.50s.
NOT verified end-to-end with a real `gt` install. Reporter has the
deterministic Graphite workflow and offered to retest. The regex +
matcher widening is a clean superset — current git-only matching still
works (verified by the 45-test smoke suite that uses `git commit` /
`git push` exclusively), and the new gt cases are pure additions.
Not in this PR:
- `gt prev` / `gt next` / `gt up` / `gt down` etc. — pure
navigation, no commit / push side effect.
- `gt restack` — could in principle rewrite commits (so the
plugin's reviewed-shas cache becomes stale), but it doesn't
create reviewable new content. Out of scope.
- `gh pr create` — already explicitly NOT a separate matcher per
the existing comment in _GIT_PUSH_RE (gh invokes git push as a
child process; the bash hook only sees the top-level
`gh pr create`). Same architectural issue as gt but with a
different cost/benefit per the existing comment.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The first round of this PR removed SKIP_WIN32, fixed venv_py to use
Scripts/python.exe, and added Lib/site-packages to the consumer glob —
all necessary. Windows verification (Win11 ARM64, Py 3.13, Git Bash)
showed two more blockers, both addressed here.
1. Pip dependency resolver picks unbuildable cryptography on ARM64.
Without --prefer-binary, pip picks a cryptography version with no
published ARM64 wheel and tries to build it from source. That needs
Rust/Cargo, almost never present on user machines → BUILD_FAILED
with err_kind=other:cryptography. A binary wheel exists for an
adjacent version (cryptography-46.0.3-cp311-abi3-win_arm64.whl);
--prefer-binary tells pip to pick it. Cross-platform safe (no-op
where the latest version already has a wheel).
2. pywin32 .pth files aren't processed by sys.path.insert().
With the venv built, ensure_agent_sdk.py's post-build probe passes
(it runs from venv_py, where Python's site.py at startup processes
pywin32.pth and registers win32/, win32/lib/ plus runs
pywin32_bootstrap.py to set the DLL search dir). But llm.py runs in
the hook's SYSTEM Python and adds the venv via sys.path.insert(),
which doesn't trigger site.py at all. Without the bootstrap, the
SDK's mcp.client.stdio → mcp.os.win32.utilities chain raises
ModuleNotFoundError: pywintypes and the agentic reviewer falls back
to single-shot silently — exactly the symptom this PR is trying to
fix. The probe says NOOP_VENV; the actual consumer fails. Probe and
consumer use different Pythons.
Replicate what site.py would do: after inserting site-packages,
also insert win32/ and win32/lib/, then exec pywin32_bootstrap.py.
Pulled into a shared helper _inject_agent_sdk_venv_into_syspath()
so both consumer sites (3P SDK fallback, agentic_review fallback)
call the same code — Windows handling stays in one place.
Verified on macOS (POSIX path unchanged):
- Helper end-to-end test: POSIX-layout venv detected + fake package
imports successfully via the injected path
- Windows-layout venv also detected; win32 branch correctly skipped
via sys.platform check
- Both files pass py_compile
Credit: @mhegazy verified the previous commit on Win11 ARM64 / Py 3.13
/ Git Bash, surfaced both issues end-to-end, and provided the exact
fix patterns. This commit applies them with the pywin32 part factored
into a shared helper (vs. inlining at both consumer sites).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The agentic reviewer is silently no-op on Windows today. SessionStart
bootstrap (ensure_agent_sdk.py) short-circuits with SKIP_WIN32 because
the consumer glob in llm.py only matches POSIX venv layout
(lib/pythonX.Y/site-packages). On Windows, venvs use Lib/site-packages
(capital L, no pythonX.Y subdir), so even if a venv got built the
glob wouldn't find its contents.
Result: Windows users on default installs (no system-wide
claude_agent_sdk) get layer 1 (pattern warnings) and layer 2 (single-
shot LLM diff review) but not layer 3 — the cross-file agentic review
that catches IDOR, auth-bypass, cross-file SSRF, and other things that
need to read related files. Plugin description claims layer 3 but it
silently doesn't run.
Three changes:
1. llm.py — extend the consumer glob (2 sites: 3P SDK fallback at
~L297, agentic_review fallback at ~L1090) to also match the Windows
Lib/site-packages layout, so a venv built on Windows is actually
discoverable.
2. ensure_agent_sdk.py — remove the sys.platform == 'win32' early-exit
so the SessionStart bootstrap builds the venv on Windows too.
Outcome code 4 (formerly SKIP_WIN32) is retired but not reused so
pre-fix telemetry rows still decode correctly.
3. ensure_agent_sdk.py — venv_py path now branches on sys.platform:
Windows venvs put the interpreter at Scripts\python.exe; POSIX
uses bin/python. Previously assumed POSIX, so even with the glob
fix, the post-build SDK-importability probe would fail on Windows.
Verified locally on macOS:
- glob test: both layouts now match (POSIX venv detected, simulated
Windows venv also detected via the new Lib/site-packages branch)
- both files pass py_compile
- POSIX path unchanged (sys.platform != 'win32' so old branch runs)
Not verified on Windows in this commit — needs an actual Windows
runner to confirm the venv build + SDK import + subprocess plumbing
all work end-to-end. The SDK spawns a child claude.exe; Windows
process plumbing has its own quirks (shell semantics, path escaping)
that may surface separately. Worth a controlled rollout (one-week
soak under env-var opt-in before flipping default).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Fixes#2043. On Git Bash for Windows, Claude Code hands script paths to
the shim in POSIX form (`/c/Users/...`). We exec a Windows `python.exe`
(the `python3` Microsoft Store stub fails the probe), and Windows Python
interprets the leading `/` as the root of the current drive — `/c/...`
becomes `C:\c\Users\...` or `D:\c\Users\...` depending on which
drive the shell happens to be on, fails with ENOENT, and every
Edit/Write/MultiEdit blocks until the session restarts.
Convert absolute path args via `cygpath -w` (a Git Bash builtin) before
exec. Guarded by `command -v cygpath` so macOS/Linux fall straight
through unchanged; `cygpath -w` is idempotent on already-Windows paths
so the rare mixed-form case is safe. Only `/*` paths are converted —
Windows-form paths reaching the shim are already openable by python.exe.
Verified locally:
- cygpath absent on macOS → guard skips → POSIX behavior unchanged
- end-to-end shim invocation with a POSIX path on macOS exits 0
- stubbed cygpath -w on /c/Users/test/hook.py produces C:\Users\test\hook.py
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Paths containing spaces (common on Windows, e.g. C:\Users\Some User\...)
cause shell word-splitting when CLAUDE_PLUGIN_ROOT is unquoted, resulting
in hooks erroring with "No such file or directory" on every tool call.
Wraps the path in double quotes for all five affected hook commands.
Fixes the pattern reported in issue #57946. Closes the fix surfaced in PR #1921.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
