Tooling: audit which PRs Codex never reviewed (quota gaps) and revisit them

[PhysShell]

Codex code review has a usage quota, and when it runs out a PR can merge without its pass (this happened on PR #39: the re-review after the left-edge fix never ran).

Task: a script (e.g. scripts/codex-review-audit.sh or a small Rust/Python tool) that walks all PRs in the repo via the GitHub API and lists the ones Codex never reviewed (no review and no inline comments from chatgpt-codex-connector[bot] on the head commit), so they can be revisited — e.g. by commenting @codex review once quota returns.

Sketch:

• list merged/open PRs;
• for each, check reviews + review comments for the Codex bot, and whether the latest commit was covered (Reviewed commit: <sha> in the bot's summary comment);
• output a table of PRs needing a visit, optionally post @codex review with a --fix flag.

Origin: session of 2026-06-11, after the Codex quota ran out on PR #39.

[PhysShell]

@codex — could you propose an implementation for this? A tool that audits which PRs the Codex bot never reviewed (quota gaps) and flags them for a @codex review revisit, following the sketch in the issue (walk PRs → check for chatgpt-codex-connector[bot] reviews/inline comments and whether the head commit was covered via the Reviewed commit: line → output the PRs needing a visit, with an optional --fix to post @codex review).

Shape and language are your call (scripts/codex-review-audit.sh or a small Rust/Python tool). Please open a PR with your proposed approach — we'd like to review the design before merging rather than merge straight away. Thanks!

---

Generated by Claude Code

[chatgpt-codex-connector]

Summary

• Added scripts/codex_review_audit.py, a standard-library Python GitHub REST API tool that paginates PR/review/comment endpoints, detects Codex bot coverage by Reviewed commit: <sha> summaries or head-commit inline/review activity, and emits a markdown table of PRs needing a revisit. scripts/codex_review_audit.pyL40-L59 scripts/codex_review_audit.pyL117-L141 scripts/codex_review_audit.pyL159-L172
• Added CLI flags for repo/token discovery, bot login override, PR state selection, showing all rows, and --fix posting of the revisit comment (@codex review) to affected PRs. scripts/codex_review_audit.pyL201-L232
• Added unit coverage for summary-line coverage, stale reviews, inline comment coverage, no-bot-activity gaps, and markdown output filtering. tests/test_codex_review_audit.pyL33-L88
• Recorded the implementation/design decision and prior art in the decisions log, including why this is a Python stdlib script rather than a Rust subcommand or gh dependency. docs/decisions.log.mdL1307-L1319
• Committed the changes on the current branch and opened a PR via the make_pr tool.

Testing

• ✅ python -m unittest discover -s tests -p 'test_codex_review_audit.py'
• ✅ python -m py_compile scripts/codex_review_audit.py
• ✅ cargo fmt --all -- --check
• ✅ cargo test --workspace
• ✅ cargo clippy --all-targets -- -D warnings

View task →