pr-watch

Polls GitHub PR CI checks and auto-fixes failures via a bounded implementer loop.

1|1|Updated Oct 3, 2021
One-click install
npx skills add https://github.com/benjr70/Smart-Smoker-V2 --skill pr-watch-benjr70
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: pr-watch
Source: https://github.com/benjr70/Smart-Smoker-V2/tree/main/.claude/skills/pr-watch
Command: npx skills add https://github.com/benjr70/Smart-Smoker-V2 --skill pr-watch-benjr70

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Newly opened pull requests often fail CI, and babysitting those checks manually wastes time. This Skill watches a PR's checks, dispatches an implementer agent to fix failures, and either lands the PR green or marks it draft with an escalation label when the fix budget runs out. ## Core Features & Use Cases - Autonomous CI watch-and-fix loop: Polls checks with a background waiter, gathers failure context (issue body, PR diff, failed job logs), spawns an implementer agent, and pushes append-only fix commits for up to 10 rounds. - Dependabot bot mode: The --bot flag switches to a 3-attempt budget shared across fires via marker comments, regenerates lockfiles with npm install --legacy-peer-deps, and applies the AFK:deps-failed label instead of AFK:checks-failed. - Safe exhaustion handling: On budget exhaustion it converts the PR to draft, applies the correct lane label, and comments for human triage — it never force-pushes, never merges, and never extends the cap. - Use Case: After /afk-pickup opens a PR on feat/issue-42, invoke pr-watch to sit on the checks; if the build fails, it automatically fixes, pushes, and re-polls until green or until 10 rounds are spent. ## Quick Start Ask the agent to run pr-watch with the PR number, branch, repo, and issue number, adding the --bot flag if it is a Dependabot pull request.

Frequently Asked Questions about pr-watch

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I automatically fix failing CI checks on a GitHub pull request?

Invoke pr-watch with the PR number, branch, repo, and issue number. It polls checks with a background waiter, gathers the issue body, PR diff, and failed job logs, then spawns an implementer agent to stage a fix which it commits and pushes, repeating up to 10 rounds.

How does pr-watch handle Dependabot pull requests differently?

Pass the --bot flag to switch to the Dependabot lane. The fix budget becomes 3 attempts total tracked via marker comments keyed to the PR head sha, fixes start by regenerating the lockfile with npm --legacy-peer-deps, and exhaustion applies the AFK:deps-failed label on the PR.

What happens when the fix round limit is exhausted?

The PR is converted to draft, labeled AFK:checks-failed (or AFK:deps-failed in bot mode), and a comment is posted requesting human triage. The skill never extends the cap, treating exhaustion as the signal to escalate to a human.

Can pr-watch merge or force-push a pull request?

No. It only pushes append-only fix commits with plain git push and never merges; green CI is the verdict while merging stays human-gated. If a push is rejected due to branch divergence, it aborts with an error for manual triage.

Why does pr-watch error out when bot marker history is unreadable?

If the head sha or PR comments cannot be read, the skill exits with an error instead of assuming zero attempts spent. This prevents a Dependabot bump that already failed three times from receiving a fresh full fix budget.