AgentStack
Browse Sign in
Browse Why AgentStack Sell Docs
Sign in
SKILL verified Apache-2.0 Self-run

Watch Pr

skill-vrallev-agent-skills-watch-pr · by vRallev

Monitor a GitHub pull request until every CI check on its current head is successful and reviewer feedback has stopped arriving. Use when the user asks to watch, monitor, babysit, or wait on a PR; poll every minute, address CI failures as they occur, invoke $address-pr-comments for review feedback, use $commit-changes for each individual CI failure fix or accepted reviewer request, push commits,…

No reviews yet
0 installs
34 views
0.0% view→install

Install

$ agentstack add skill-vrallev-agent-skills-watch-pr

✓ scanned · ✓ verified, works with Claude Code, Cursor, and more.

Security review

✓ Passed

No issues found. Passed automated security review. · v0.1.0 How review works →

  • Prompt-injection patterns
  • Secret / credential exfiltration
  • Dangerous shell & filesystem operations
  • Untrusted network calls
  • Known-malicious package signatures

What it can access

  • Network access No
  • Filesystem access No
  • Shell / process execution No
  • Environment & secrets No
  • Dynamic code execution No

From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.

View the full security report →

Verified badge

Passed review? Show it. Paste this badge into your README, it links to the public security report.

AgentStack Verified badge Links to your public security report.
[![AgentStack Verified](https://agentstack.voostack.com/badges/verified.svg)](https://agentstack.voostack.com/security/report/skill-vrallev-agent-skills-watch-pr)

Reliability & compatibility

Security review passed
0 installs to date
no reviews yet
2mo ago

Declared compatibility

Claude CodeClaude Desktop

Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.

Preview Execution monitoring

We're building live execution health for every listing: tool-call success rate, median latency, uptime, and last-checked timestamps, measured, not self-reported. It isn't live yet, so we don't show numbers we can't stand behind.

How agent discovery & health will work →
Are you the author of Watch Pr? Claim this listing to set pricing, connect Stripe payouts, and keep 70% of every sale.
Sign up to claim

About

Watch PR

Monitor one pull request through CI and review feedback. Do not merge the PR.

Setup

  1. Resolve the supplied PR URL or number, repository, head branch, and current head SHA. If no PR is supplied, resolve the open PR for the current branch; ask only if the result is missing or ambiguous.
  2. Confirm write access to the PR branch and use a checkout whose current branch maps to that PR. Preserve unrelated local changes; use a separate worktree when switching would disturb them.
  3. Prefer thread-aware GitHub connector reads. Use gh when connector coverage is insufficient and authentication permits it.
  4. Take an initial machine-readable snapshot of:
  • every CI check and job for the current head SHA,
  • inline review threads and all comments in each thread,
  • PR reviews and PR-level conversation comments.
  1. Record stable comment identifiers or timestamps so later snapshots detect newly posted feedback, including replies to previously handled threads.

Monitoring Loop

Use the product's recurring monitoring or wait mechanism when available. Otherwise poll every minute without a tight shell loop.

On every poll:

  1. Refresh PR state and the head SHA. If the head changed, discard the old CI result and restart green/quiet confirmation for the new head.
  2. Compare all review and PR-comment streams with the previous snapshot.
  3. When new human-authored feedback exists whose latest reply is not an existing **Ralf-AI:** response, run $address-pr-comments from /Users/ralf/.codex/skills/address-pr-comments/SKILL.md for the PR. Follow all of its safety, commit, validation, push, reply-prefix, and unresolved-thread rules, with the classification rules below.
  4. After $address-pr-comments finishes, refresh the head SHA, CI, and comments immediately. Any pushed commit or new comment resets green/quiet confirmation.
  5. Continue waiting while a current-head CI job is missing, queued, pending, in progress, stale, or has any conclusion other than success. Do not count skipped, neutral, cancelled, timed-out, or action-required jobs as successful.
  6. Address CI failures as they occur. Inspect each failing check or job enough to identify the concrete failure, implement the smallest appropriate fix for that individual failure, run focused validation, then use $commit-changes from /Users/ralf/.codex/skills/commit-changes/SKILL.md to create one commit for that failure before pushing. Keep each individual CI failure fix in its own commit.
  7. If a CI failure is transient, infrastructure-owned, externally blocked, unrelated to the PR, or unsafe to fix without more context, report the blocker and continue monitoring instead of making a speculative change.

Feedback Classification

Apply these additions while using $address-pr-comments:

  • Reasonable suggestion: Accept a clear, safe request that improves correctness, reliability, tests, readability, or maintainability without materially expanding the PR. Make one new commit for that request using $commit-changes, run focused validation, push it, and reply with the commit SHA.
  • Unreasonable suggestion: Reject only when the request is clearly incorrect, irrelevant, duplicative, contrary to verified project constraints, or a disproportionate scope expansion. Make no code change. Post a concise, respectful GitHub reply with the concrete reason and prefix it exactly with **Ralf-AI:**.
  • Ambiguous, conflicting, or risky suggestion: Do not label uncertainty as unreasonable. Ask the user before editing or posting a speculative answer, as required by $address-pr-comments.
  • Question: Answer directly on GitHub when the answer is verified. Do not create a commit for a question-only response.

Never amend or force-push. Keep one commit per accepted reviewer request. Never resolve a review conversation unless the user explicitly asks.

Completion

Finish only when all of these are true for the same current head SHA:

  • At least one CI check has been discovered, every discovered CI check and job has completed with conclusion success, and no expected or required check is missing.
  • Every observed human review or PR-level comment has been considered by $address-pr-comments; the latest reply in each handled conversation is **Ralf-AI:** or no reply was needed.
  • Two consecutive full snapshots, at least one minute apart, have the same head SHA and no new human comments, while CI remains fully successful.

If the PR is merged or closed, stop and report that terminal state. If authentication, branch permissions, an ambiguous/risky request, or a permanently failed external check requires owner action, report the exact blocker and required action; do not falsely declare the PR green.

Give a concise final handoff with the final head SHA, CI result, handled and rejected feedback, commits pushed, validation run, replies posted, and any remaining manual action.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

Install and usage instructions live in the source repository linked above.

Reviews

No reviews yet, be the first.

Versions

  • v0.1.0 Imported from the upstream source.