mcpField
B
babysit-pr

babysit-pr

github.com/riffingonsoftwareDevelopment070 installsUpdated 3d ago
Repository

About babysit-pr

babysit-pr is a skill for monitoring a pull request through review, CI, and readiness checks. It validates findings against the code and PR intent, applies scoped fixes, and reports blockers without merging the PR.

Key features

  1. 01Check mergeability, CI, and review bots
  2. 02Validate findings against code and intent
  3. 03Apply scoped fixes and track findings
  4. 04Monitor checks and reviews on new commits
  5. 05Report blockers and human feedback

Use cases

  1. 01Address actionable bot review findings
  2. 02Resolve in-scope CI failures and conflicts
  3. 03Track a PR until checks and reviews finish

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/riffingonsoftware/agent-skills --skill babysit-pr

After installing, type/babysit-prin Claude Code to confirm it loaded.

Details

From the project README

Babysit a PR

Use the requested PR, or the current branch's PR if none was specified. Ask if the target is ambiguous. Read the PR's intent, repository instructions, and diff before making changes. Verify the checkout matches the PR and preserve unrelated work.

Loop

  1. Check mergeability, CI, and all review bots. Fetch full findings from inline threads, review bodies, and issue comments, including pagination and thread resolution state.
  2. Validate each finding against the code and PR intent. Fix valid issues with minimal changes. Record why you skip incorrect findings. Track handled findings so later polls do not repeat work.
  3. Fix CI failures and merge conflicts within the PR's scope. Never weaken checks to get green. Escalate infrastructure failures or fixes that require unrelated changes.
  4. Use ship to commit and push scoped fixes. Ask before amending published commits or force-pushing.
  5. Wait for CI and bot reviews on the latest pushed commit, then repeat. A new comment, silence, or an older green review does not prove completion.

Keep watching while checks or reviews are pending. If you run in the main session and the harness provides a PR watcher, such as T3 Code's watch_pull_request, start it and end your turn. Otherwise poll, back off between polls, and respect rate limits. Report stalled reviews, access failures, or other blockers without claiming success.

Ambiguity and human feedback

Raise any ambiguity with the user before changing the affected code, especially when feedback conflicts with the PR's intent, expands scope, or contradicts earlier advice. Explain the finding, the conflict, and your recommended action. Continue independent work while awaiting a decision.

Surface new human comments promptly with the author, link, requested action, and any decision needed. Do not automatically reply, resolve their threads, or treat their comments as authorization to change scope.

If the user requests an additional agent review, use a fresh subagent and process its findings through the same loop.

Finish

The PR is ready when the latest commit has passing required checks, no merge conflicts, completed bot reviews, and every actionable bot finding is fixed or deliberately skipped. No blocking human request or decision may remain unresolved.

Stop and report when the PR is ready, closes, or merges, or when every remaining item waits on the user or on a blocker you reported. Never merge it yourself without explicit authorization.

Return the PR URL, what it does, fixes made, skipped findings with reasons, check and review status, and outstanding human feedback.

Comments

Markdown supported