fix(formatters): PATH probe notices + monorepo markdownlint walk (#2732)

[kyle-sexton]

Summary

Fleet inventory of formatter/lint command -v probes (docs/formatter-path-probes.md). Sibling hooks now emit accurate degraded notices (this-edit / notice-only latch / environment inheritance / PATH probed:) without nvm layout guessing. markdown-format walks from the edited file up to $REPO_ROOT for workspace node_modules/.bin installs.

Test plan

• plugins/markdown-format/hooks/markdown-format.test.sh — 149/149 (includes monorepo walk)
• plugins/bash-format/hooks/bash-format.test.sh — PATH diagnostic assert
• plugins/actionlint/hooks/actionlint-check.test.sh — minimal-PATH absent notice

Related

Closes #2732

[cursor]

Bugbot is not enabled for your account, so this pull request was not reviewed.

Enable Bugbot in the Cursor dashboard to get automatic reviews on future PRs.

[claude]

Claude finished @kyle-sexton's task in 4m 33s —— View job

---

Security review (/review:security-review)

Reviewed the diff against origin/main (git diff origin/main...HEAD, HEAD ce5f44a). Scope: 25 changed files — 24 are notice-text/version-bump/test-only changes (the new PATH probed: ${PATH} diagnostic lines echo the hook's own inherited PATH back into its own advisory notice; not attacker-influenced input and not a meaningful disclosure). The one file with real logic changes is plugins/markdown-format/hooks/markdown-format.sh's resolve_repo_markdownlint(), rewritten to walk from the edited file up to $REPO_ROOT for monorepo node_modules/.bin installs.

1 finding posted inline (severity: IMPORTANT):

• plugins/markdown-format/hooks/markdown-format.sh#L466-L492 — permalink. The refactor changed the symlink-follow loop's three failure branches (depth-limit exceeded, readlink failure, missing cygpath for a Windows-style link target) from return 1 to break. Because the function always returns the original unresolved $candidate path (not the manually-walked $target), and the kernel independently fully resolves symlinks at exec time (Linux ELOOP ≈ 40, past this loop's 32-hop cap), a 33-hop relative symlink chain planted under node_modules/.bin/ can pass this function's own trust-boundary validation while its final, unread hop actually resolves outside the repository — defeating the exact protection the function's header comment (L448-453) describes ("rejects a checked-in or replaced .bin symlink that escapes the repository trust boundary"). No test in markdown-format.test.sh covers resolve_repo_markdownlint's own binary-resolution symlink chain (only the edited-file-escape case and the new monorepo happy path are tested). Full exploit chain and a suggested fix (restore return 1 for these three branches) are in the inline comment.

No other security-relevant findings — no injection, auth/access-control, secrets handling, or GitHub Actions permission/trigger issues in this diff (the .claude-plugin/plugin.json and CHANGELOG.md changes are version-bump/doc only).
· branch cursor/2732-formatter-path-8f91

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: ce5f44ad2d

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[github-actions]

Last security-reviewed head: ce5f44ad2d6ffdb5c8fbedf50da178f0785c9357. On the next push, the relevance gate compares only the commits since this SHA; delete this comment to force a full re-review.

[claude]

Claude finished @cursor[bot]'s task in 2m 38s —— View job

---

Code review (/review:code-review)

• Gather context (PR 2801, head da8cd42e)
• Determine actual PR diff scope
• Review changed files
• Post findings

Skip gate hit: PR is already merged.

PR 2801 is in the MERGED state, which is condition 1 of the code-review skill's skip gate ("PR is closed or not open") — per the skill, deep review stops here and nothing further is posted.

For context on what I verified before stopping: I diffed head da8cd42e against origin/main and confirmed every file this PR touches (docs/formatter-path-probes.md, and the markdown-format/actionlint/bash-format/biome-format/go-format/ruff-format/typos-format plugin files) is byte-identical to what's already on main — including resolve_repo_markdownlint() in plugins/markdown-format/hooks/markdown-format.sh, where the three symlink-follow failure branches already use return 1 (not the break that the earlier Codex and Claude reviews on this PR flagged as a trust-boundary bypass). That fix is present on origin/main already, so there's nothing outstanding to re-review or re-flag.

[github-actions]

Claude has reviewed this PR 1 time. The lane skips further automatic reviews after 5; deleting this comment resets the count.