Skip to content
Get packs

Guide · Agent skill

How to use the Babysit PR agent skill: install and trigger it

Updated · written with AI from the project's own documentation and checked against our catalogue data

Babysit is an agent skill that keeps an AI coding agent working on a GitHub pull request until it is ready to merge, instead of checking it once and stopping. It is written for developers and teams who already use GitHub and want an agent to handle the repeated cycle of checks, review comments, fixes and pushes. After this guide you will know what the skill does, how to install it, how to ask for it in plain language, and what it needs and touches on your machine.

Babysit PRthedotmack/claude-mem · Apache-2.0

Quick start

  1. 1Download and unzip the skill. Download the ZIP from this page and unzip it so the babysit folder containing SKILL.md is in place.
  2. 2Put the skill in the right folder. Copy it to ~/.claude/skills/ for personal use, or to .claude/skills/ in a project. In the Claude apps, upload it under Settings → Capabilities → Skills instead.
  3. 3Install and sign in to the requirements. Make sure gh, jq and git are installed, and that gh is authenticated with access to the repository.
  4. 4Check out the PR branch locally. Switch your local repository to the pull request's branch so the agent can edit code and run tests.
  5. 5Ask the agent to babysit the PR. Send a plain request such as "Babysit PR 123 until it's ready to merge", including the PR number if you know it.
  6. 6Review the final evidence. When the agent reports back, check the latest commit SHA, check results, unresolved thread count and any untouched dirty files.

What is the babysit skill and what does it make the agent do?

An agent skill is a folder containing a SKILL.md file with instructions that an agent loads when a request matches the skill's description. The babysit skill (plugin/skills/babysit from thedotmack/claude-mem, licensed Apache-2.0) is for one job: watching a pull request or review cycle until it is ready to merge.

When asked to babysit a PR, the agent identifies the PR number, branch and base branch, confirms the PR is not a draft, and checks mergeability, CI checks, review decision, comments and review threads. It polls pending checks, reads new comments, verifies whether bot or human feedback points to real problems, fixes genuine issues in focused commits, runs relevant tests or builds, pushes, and repeats.

  • It resolves stale review threads only after confirming the code now addresses the comment.
  • It stops only when checks pass or are intentionally skipped, the review decision is acceptable, and no actionable comments or unresolved threads remain.
  • Before reporting, it does a fresh sweep and reports evidence such as the latest commit SHA, check results, unresolved thread count, tests run, and any dirty local files left untouched.

Install the babysit skill from the ZIP or the Claude apps

Download the ZIP from this page and unzip it. For a personal install, place the skill folder in ~/.claude/skills/. To make it available only inside one project, place it in that project's .claude/skills/ folder.

If you use the Claude apps instead, upload the skill under Settings → Capabilities → Skills.

  • Personal install: ~/.claude/skills/
  • Project install: .claude/skills/ inside your repository
  • Claude apps: Settings → Capabilities → Skills, then upload

Check the requirements before you start

The skill depends on several command-line tools and a specific setup. Confirm each one is available before asking the agent to babysit a PR.

The skill needs the GitHub CLI (gh), authenticated with access to the repository, plus jq for parsing JSON output, git, and a local checkout of the PR branch. It uses GitHub's GraphQL API for review threads, so it is not suited to projects hosted elsewhere.

  • GitHub CLI (gh), authenticated with access to the repository
  • jq, used to parse JSON output
  • git
  • A local checkout of the PR branch
  • Network access, because it calls GitHub

Trigger the skill with natural requests

You do not need special syntax. The skill's description covers requests to babysit, monitor, or keep checking PR comments, reviews and CI until actionable issues are resolved. Say what you want in plain language and include the PR number when you have it.

The agent will still work through the workflow in order, so a request like this is enough to start a full review-and-CI loop.

  • "Babysit PR 123 until it's ready to merge."
  • "Keep checking the comments and CI on this pull request and fix anything real."
  • "Monitor the reviews on my current branch's PR until all threads are resolved."
  • "Watch CI on PR 456 and tell me when it's clean."

How the agent checks status, threads and CI

The skill's instructions show the exact GitHub CLI calls it uses. You can read them in SKILL.md to see what will run. Coarse status comes from gh pr view with JSON fields such as mergeable, reviewDecision and statusCheckRollup. Review threads come from a GraphQL query that pages through results with pageInfo and filters for unresolved threads using jq.

Polling happens at a practical interval, usually 30 to 60 seconds, unless you ask for a different cadence. The instructions say to keep the watcher running while long checks are pending.

  • gh pr view <number> returns status fields for the PR
  • gh api graphql fetches review threads, paging with the end cursor when more pages exist
  • Resolving a thread uses a GraphQL mutation, and only after the fix is verified
gh pr view <number> --json \ number,state,isDraft,mergeable,mergeStateStatus,reviewDecision,headRefOid,statusCheckRollup,url

Safety notes: scripts, network and pushes

This skill runs scripts and uses the network. It calls GitHub through gh and GraphQL, and it can commit and push changes to the PR branch and resolve review threads. The source says it does both automatically once fixes are verified, so it is not a read-only tool.

Read SKILL.md before installing, confirm the gh account has only the access you intend, and make sure the local checkout is the branch you expect. The skill also reports any dirty local files it leaves untouched, which you should review.

  • Runs scripts and makes network calls to GitHub
  • Can push commits to the PR branch
  • Can resolve review threads once fixes are verified
  • Use an account and checkout you are comfortable letting the agent change

Tips for getting a clean result

Start with a PR that has a local checkout on its branch, so the agent can fix code and run tests. If your repository has many review threads, the skill pages through them, so you do not need to split the work.

If a bot reports a problem against old code, the skill is designed to check whether the thread is outdated or already addressed in the latest head. If your project ships generated files, the instructions say to confirm the source and generated artifact agree before resolving comments.

  • Ask for a different polling interval if 30 to 60 seconds does not suit your CI.
  • Check the final report for the commit SHA, check results and unresolved thread count.
  • Keep a clean working tree before you start, so unrelated local changes are not mixed in.

What people use it for

  • Shepherding a PR through automated reviewer comments from bots until no actionable threads remain.
  • Watching CI on a pull request and fixing failing tests or builds as they appear.
  • Re-checking a PR after a push so new review comments are read and addressed without manual polling.
  • Confirming that a stale review comment is outdated or already fixed before resolving its thread.

Frequently asked questions

Is the babysit skill free to use?

The source gives the licence as Apache-2.0 for the thedotmack/claude-mem repository, which contains this skill. It does not describe pricing, so check the repository's licence file for terms.

What do I need installed to use babysit?

You need the GitHub CLI (gh) authenticated with access to the repository, jq, git, and a local checkout of the PR branch. The skill also needs network access to GitHub.

Where do I install the babysit skill?

Unzip the downloaded ZIP into ~/.claude/skills/ for personal use or into a project's .claude/skills/ folder. In the Claude apps, you can upload it under Settings → Capabilities → Skills.

Does babysit work with GitHub Enterprise or GitLab?

The source says the skill depends on the GitHub CLI and GitHub's GraphQL API, and it is not meant for projects not hosted on GitHub. It does not describe other hosts, so do not assume it works with GitLab or Bitbucket.

Will babysit push code or resolve comments without asking?

Yes, according to the source. Once fixes are verified, the skill pushes changes and resolves stale review threads automatically. If you need approval for each step, do not use this skill.

How is babysit different from just asking an agent to check my PR once?

A single check stops after one look. Babysit keeps polling checks, reading new comments, fixing real issues and pushing until the PR is clean, then reports evidence of its final state.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Babysit suits developers and teams on GitHub who want an agent to carry a pull request through review comments and CI without repeated manual checks. Skip it if you need a person to approve every commit or thread resolution, or if your project is not hosted on GitHub.

Download the skill