[kyle-sexton]

No related issue: rolls out the guardrails lefthook remote (claude-code-plugins#9504)

Summary

Runs the guardrails plugin's three git-hook entries in every clone of this repository from one committed lefthook remotes: entry, so each machine gets the same hooks from lefthook install and picks up upstream fixes automatically:

• commit-msg-convention (commit-msg): the subject against the rule in docs/conventions/source-control/commit-convention.yml.
• pre-commit-content (pre-commit): secrets and machine-specific paths on added lines.
• reference-verify (pre-push): CLI flags, /plugin:skill references and deleted paths on pushed lines.

Fix

• lefthook.yml (new; the repository had no hook manager): one remotes: entry, git_url: https://github.com/melodic-software/claude-code-plugins, ref: main, refetch_frequency: 24h, config plugins/guardrails/git-hooks/lefthook-remote.yml.
• That is the whole diff against main. The earlier revision of this PR vendored copies under .githooks/guardrails/ and needed ShellCheck, shfmt and machine-path exclusions for them in .github/workflows/pr-require-checks.yml, plus a subject_pattern in .claude/source-control.md; all of that is gone. The subject rule now comes from the standards-synced commit-convention.yml.
• No other changes kept.

Verification

• install.mjs verify --repo <worktree> (rc 0):

  ok	commit-msg-convention
  ok	pre-commit-content
  ok	reference-verify
  

• install.mjs plan --scope remote afterwards: no decisions left.
• lefthook install: sync hooks: ✔️(commit-msg, pre-push, pre-commit). This PR's commits ran the remote pre-commit-content and commit-msg-convention, and its push ran reference-verify.

Observation runs from the earlier revision, still valid because the remote runs the same hook scripts:

Hook	Observation	Findings	Classification
reference-verify	pre-push replay of origin/main~200..origin/main, 28 changed .md/.sh/.ps1 files, 6 s	0	none; no file in the range exceeds the argument-size limit (largest README.md, 31,947 bytes)
pre-commit-content	each of the last 100 non-merge commits replayed into the index (1 to 65 staged files each; every count matches that commit's own change list, renames counted once), 10 s total	0	none. A planted /home/<name>/... probe was caught (rc 1), so the replay exercises the check
commit-msg-convention	subject rule tested against the last 500 subjects on main (all 308)	1 non-matching: Initial commit	narrowly justified exception: GitHub's repository-creation commit, never made through a hook

Related

• claude-code-plugins#9504: the lefthook remote config this points at.
• claude-code-plugins#9426: rollout tracking and known hook defects.
• standards#852: the pilot.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	🔄 Running since 2026-10-24T04:02:10.955663Z	997186e	Draft marked ready

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" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[github-actions]

claude-review reviewed 997186ed897ad1802ead7ff80f1065c24b7063fd: 0 finding(s), 0 as inline comments.

Code review: no findings.

I read the full diff, which is one new file, lefthook.yml. It declares a single lefthook remotes: entry pointing at claude-code-plugins (ref: main, refetch_frequency: 24h, config plugins/guardrails/git-hooks/lefthook-remote.yml). I also read the PR context and the one path-scoped rule, .claude/rules/pr-body-contract.md.

• PR body: It opens with No related issue: <reason> and has ## Summary, ## Fix, ## Verification and ## Related, so it meets the PR body contract. There is no closing issue, so no acceptance criteria apply.
• Commit convention: docs/conventions/source-control/commit-convention.yml, which the PR says the commit-msg hook reads, exists in the repo and is a managed file. I confirmed the file is present; I did not run the hooks.
• Verification claims: The install.mjs verify, lefthook install and replay results in the PR are author-claimed. I did not run them and don't count them as independently confirmed.
• Unpinned remote (not reported here): The remote follows the mutable ref: main and refetches every 24h. Hook code from another repo therefore runs in every clone. I left this out of this lane because the repo has a separate security-review workflow (pr-review-security-hosted.yml), which should judge it.
• Hold classes: None apply. The change adds no model spend, doesn't touch trusted-actors, CODEOWNERS or .gitmodules, and has no critical security finding for this lane.
• Plugin manifests: The diff changes no manifests, so no manifest <field> lines apply.

[github-actions]

claude-security-review reviewed 997186ed897ad1802ead7ff80f1065c24b7063fd: 1 finding(s), 1 as inline comments.

Security review: PR 493 (head 997186e)

The diff adds one file, lefthook.yml. It declares a lefthook remotes: entry that pulls the guardrails git-hook config from melodic-software/claude-code-plugins. No Actions workflow, plugin manifest, or instruction-surface file changed, and the change touches no trust-gate path.

Finding (1, SUGGESTION): lefthook.yml lines 2-3. The remote uses the floating ref main and re-fetches every 24h. Lefthook runs whatever hook commands that remote config defines on each contributor's machine. A malicious or mistaken commit to claude-code-plugins main would therefore run on every clone at the next commit or push, with no review in this repository. The PR describes the auto-update as intended, so I raised it as a trade-off. Pinning ref to a tag or SHA and bumping it by PR would narrow it, as would confirming that branch protection and required review on the upstream main is the accepted control. I inspected only this diff and did not read the upstream hook scripts, so I did not assess what they do.

Nothing else turned up. There is no CRITICAL finding, so no hold is needed.

critical-findings: 0

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 1479f956f9

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".