[derrickaw]

Also, add /gemini review and /summarize-failures actions to documentation.

[damccorm]

Instead of doing all this bespoke context gathering, have you tried just prompting Gemini and having it do the work? Something like:

Review the CI failures in PR XYZ. Summarize the failures and print them in format ...

This is similar to how the review bot works, and I empirically found that this worked better than trying to hydrate the correct context. It is more expensive, but the token count just won't be that high regardless in my experience. I am not sure if that holds here where parsing logs can be quite a pain

[derrickaw]

Ok, tried to improve it. Thanks

[derrickaw]

/gemini review

[github-actions]

📋 Review Summary

This PR adds a new PR Failure Summary GitHub Actions workflow (pr_failure_summary.yml) triggered via /summarize-failures comments or workflow_dispatch to cluster failing CI checks using Gemini CLI and post a sticky root-cause summary comment. It also updates pr-bot and contributor documentation to surface /summarize-failures and /gemini review.

🔍 General Feedback

• The security model for the issue_comment trigger is well thought out: PR code is not checked out (ref: 'master' with sparse checkout only for setup-gh-cli-linux), Gemini CLI is restricted to an explicit read-only tool allowlist, and untrusted input is passed via files/env vars rather than interpolated into the shell script.
• Moving concurrency from the workflow level to the job level in pr_failure_summary.yml is recommended so unrelated PR comments do not cancel an in-progress failure summary run.

---

💡 Tip: Comment /gemini review on this pull request to trigger a new Gemini code review.