[abruption]

Summary

Default the contributor template to non-closing references and make intentional completion a deliberate choice. Replace frozen legacy CLI/nonexistent-document checks with the current CLI, generator, existing documentation, scoped subsystem validation and explicit environment/not-run evidence.

Add a warning-only review aid for nearby negation or partial-completion wording around closing references. It reads a bounded event JSON file as data and emits fixed annotations/summary text. The main-only pull_request_target workflow checks out its trusted github.sha, never PR head code, with read-only contents permission and no persisted checkout credentials or issue writes.

Refs #349

Validation

• Python 3.9.6, macOS arm64: 12 executable fixtures and 81 documentation tests pass; generator consistency, diff check and YAML parsing pass.
• Fixtures cover ordinary/intentional references, negation/partial phrasing, colon/case/wrapped prose/cross-repository references, multiple references, edited bodies, unavailable/oversized input, bounded regex behavior, malicious body text and summary failures.
• Root independently reran the documentation suite after the implementation handoff. Post-review regressions also check job-level write permissions and a known trailing-qualification miss.

Review and limits

Pre-PR independent review ran Claude Code 2.1.312 (requested and response-confirmed claude-opus-5-5) and Codex CLI 0.165.0 (isolated default model; actual model ID not exposed), in parallel without tools, writes, tests or networking, on frozen input SHA-256 23055c106a46819b7db872848d3c931e7d64362fad9542bc445c449ede0b0998. Codex found no mandatory code correction; Claude requested clearer limitations. The resulting template/docstring clarification and stronger permission/known-miss regressions were applied and retested afterward; the initial review is not described as a second review of identical final bytes.

This heuristic cannot determine intent or reproduce GitHub's complete closing behavior. In particular, trailing qualifications, manual sidebar links and commit-message references remain manual review concerns. It warns without blocking legitimate intentional completion or changing any issue state. GitHub's documented closing-reference rules remain authoritative.

The newly introduced production event workflow will first be available after merge into main; candidate fixtures execute the actual checker now. No real-host/native-agent check is implied by this contributor change. This PR does not merge, publish or change repository protection settings.

[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	✅ Completed 2026-10-29T13:19:54.970829Z	fa0d456	PR opened
🔒 Security Review	✅ Completed 2026-10-29T13:20:14.591937Z	fa0d456	PR opened

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.

[abruption]

Current-head verification

Reviewed head: fa0d456ded99e0a71dfd47c216e390b447ac292a. All 32 checks succeeded, including CI 40880732813. The PR is Ready.

Automatic code and security reviews both report completed at this exact head. Code findings flags are false; security finding flags are null/unavailable, so they are not represented as an affirmative false result. No review threads are posted.

12 executable fixtures and 81 documentation tests pass. The newly introduced trusted-main PR-body event workflow is not claimed as an observed production event before merge.

GitHub closingIssuesReferences is empty: this PR references issue #349 without scheduling automatic issue closure. No merge, tag, publication or operational change was performed.