What does this skill do, and when should you use it?
This skill watches an open GitHub PR over time, handling review feedback, fixing CI failures, syncing the branch, and judging merge-readiness. It is designed for long-running operation, continuing to process new event streams without stopping at the first item that needs a human. It supports multiple execution modes — self-sustaining in-session watch, checkpoint, and pipeline — to fit different automation environments.
Runs a loop that periodically fetches PR state, review threads, CI checks, and branch sync status (using the GitHub CLI and a Python script). It delegates comment resolution to ce-resolve-pr-feedback, CI debugging to ce-debug, and performs bounded routine branch updates when authorized. It maintains a state directory for progress tracking and deduplication, and decides whether to continue to other PRs in a stack or merge based on the configured posture (target, stack-ready, stack-land).
- A developer wants to keep a feature PR merge-ready without manually polling for comments and CI status.
- A CI engineer needs to automatically fix failing checks and push commits to the PR until all checks pass.
- A team managing a stack of dependent PRs wants changes to propagate up the stack automatically while keeping a single semantic writer.
- An automation pipeline (such as `lfg`) needs to run non-interactively until the PR merges or a budget is hit.
How do you install this skill?
- The skill relies entirely on GitHub API and external tools; mainland-China users may experience connectivity issues due to network restrictions.
- Automated background watching may run for extended periods and consume resources; explicit user authorization and monitoring are required.
- Static review cannot verify actual execution; testing in a real environment is recommended.
- Shell / CLI
- Network access
- Local filesystem
Python 3GitHub CLIgit
This skill is part of the compound-engineering-plugin. Install the plugin following the README instructions for your editor (Claude Code: /plugin marketplace add EveryInc/compound-engineering-plugin then /plugin install compound-engineering; Cursor, Codex, etc. are also supported). The skill is located at skills/ce-babysit-pr/SKILL.md.
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/EveryInc/compound-engineering-plugin.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/skills/ce-babysit-pr" ~/.claude/skills/
rm -rf "$tmp"Generated from the source repository and skill path; it copies only this skill's folder. If the author's install steps above differ, follow those first. To scope it to one project, replace ~/.claude/skills with that project's .claude/skills.
How do you use this skill?
Invoke it from a supported environment (e.g., Claude Code or Codex) with a prompt like /ce-babysit-pr <PR-number> [watch] [duration] [posture:target]. It requires a local git checkout, the GitHub CLI, and Python 3. Follow the skill's instructions; the watch will continue until the PR is merged/closed, a budget is reached, or the user stops it.
What are this skill's strengths and limitations?
- Continuously processes multiple event streams instead of stopping at the first blocker
- Delegates comment and CI work to specialized skills, keeping responsibilities separate
- Supports multiple automation modes (in-session, checkpoint, pipeline, durable scheduler)
- Persists state between runs for clean resume
- Clear boundaries: never merges (except stack-land), never operates on drafts unless explicitly requested
- GitHub-only; does not support GitLab or other forges
- Requires local git checkout, GitHub CLI, and Python 3; sandboxed GUI environments can't run it
- Needs a background-and-wake capability (like Claude Code's Bash+Monitor) for fully autonomous monitoring, otherwise falls back to checkpoint mode
- May auto-modify multiple PRs in a managed stack, requiring confirmed stack membership
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| PR Babysitter this page | 52 · Use with care | ★ 25k | 3d ago | MIT |
| PR Babysitter: Watch Pull Requests Until Merge | 47 · Use with care | ★ 98k | 3d ago | Apache-2.0 |
| PR Babysitter ✓ OpenAI · Official | 57 · Use with care | ★ 128k | 3d ago | Apache-2.0 |
| GitHub PR Workflow Assistant | 42 · Not recommended | ★ 1.7k | 3d ago | MIT |
| Session Pull Request Assistant ✓ Microsoft · Official | 43 · Not recommended | ★ 193k | 3d ago | MIT |
How did FollowSkills review this skill?
The skill explicitly restricts to GitHub, defines a clear mutation envelope (no merge, no force-push, no stack structure changes), and emphasizes explicit user authorization for cross-PR actions. However, no dependency security checks or data-flow transparency details are provided, and some automated operations (e.g., background watching) may lack user confirmation.
The skill is detailed and stepwise with a clear state machine and dedup protocol. However, static review cannot verify the executability of key paths, and reliance on external tools (e.g., gh, git, pr-snapshot) and network access may affect reliability; failure feedback mechanisms are not fully demonstrated.
The scope is clear (GitHub PR monitoring), triggers are specific, and boundaries are well-defined (not a one-shot request). However, the skill primarily depends on GitHub API and may not be suitable for non-GitHub environments; no mention of network reachability, especially for mainland-China users where GitHub access may be restricted.
The documentation is well-structured with detailed security notes and state definitions, but lacks FAQ, changelog, and version numbering. Install/dependency notes are incomplete, and maintenance responsibility is not explicit.
The skill's goals and expected outputs are clear, but actual completion cannot be confirmed from static source. Output format and direct usability are unverified, and autonomous monitoring may incur additional costs.
The repository includes CI configuration and tests, but they do not cover the full key paths of the skill, and no independent third-party verification evidence exists. Static review limits the depth of verification.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
See the full review method →