PR readiness

SkillDev tools

Validate changes and pull requests through merge readiness, including local gates, Codex review, CI, independent review, manual evidence, and unresolved review threads.

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the PR readiness skill

What this skill tells your AI

The instructions your AI receives, as published by christitustech/titus-ai in .agents/skills/pr-readiness/SKILL.md and read by ahel’s review.

Workflow

  1. Read the applicable repository instructions, requirements, accepted plan, and current task status.

  2. Inspect the branch, worktree, untracked files, and complete diff. Separate unrelated pre-existing changes from the requested change.

  3. Run focused checks, then the repository's complete required local gate. Record exact commands, results, skipped checks, and residual risk.

  4. Run the built-in local Codex review after the complete local gate passes:

    codex review --uncommitted
    

    Let the review finish. Verify every finding against the current diff, fix only actionable defects, add regression coverage when practical, rerun affected validation, and repeat codex review --uncommitted until no actionable findings remain. Explain verified false positives without changing correct code. A review-mode Codex instance must report findings directly and never launch a nested review. If Codex review is unavailable or fails, report it as a readiness blocker instead of substituting a third-party review CLI.

  5. Commit, push, or open a pull request only when the user authorized those state changes. Open a ready-for-review pull request by default. Create or leave it as a draft only when the user explicitly requests a draft. If known gates or required manual tests remain, stop before opening the pull request and report the blockers instead of using draft state as a holding area.

  6. For a published PR, verify checks and reviews against the latest commit. Inspect thread-level resolution state rather than relying only on flat comments. For a contributor-fork PR, read references/contributor-forks.md.

  7. Require a fresh independent review. The builder's self-review and a green CI run do not replace it.

  8. Complete and document the repository's manual-test checklist on the real target environment when practical.

  9. Recheck the final diff, required checks, reviews, and unresolved threads after every push.

Hard gates

Do not report a pull request as ready to merge while any of these remain:

  • required CI is failed, pending, missing, or attached to an older commit
  • actionable review feedback is unresolved
  • the Codex review loop or a required independent review is incomplete
  • required manual testing is incomplete or undocumented
  • the branch contains unrelated changes, secrets, debug code, or generated junk
  • planning documents no longer match the implementation

Do not perform the merge unless the user explicitly requests it. Never fabricate commit hashes, review state, commands, or test results. For uncommitted work, report HEAD together with worktree and untracked-file state. When no pull request exists, report remote checks, reviews, threads, and PR status as not applicable.

Pull request evidence

Ensure the PR records:

  • the problem and implemented approach
  • important decisions and deviations from the plan
  • exact automated checks that passed
  • manual tests and their environment
  • screenshots or recordings for visible changes
  • known limitations, skipped validation, and follow-up work

Final report

Report HEAD, worktree and untracked-file state, changed-file scope, local checks, manual-test state, and remaining blockers. For a published pull request, also report its latest commit, remote checks, review state, unresolved threads, and whether it is draft, review-ready, or merge-ready.

Signals

GitHub stars
125
Forks
21
Last commit
Sep 2026
Advanced
Catalog kind
skill
Gateway key
pr-readiness
Source
github.com/christitustech/titus-ai