Guide · Agent skill
How to use the Post-Release Babysitter agent skill (childhood)
Updated · written with AI from the project's own documentation and checked against our catalogue data
The childhood skill is an agent skill for the claude-mem project that tells an AI agent how to keep checking a release after it ships. It covers the published package, the live production service, the plugin installed on a user's machine, and late CI results or user reports. This guide explains what the skill does, how to install it, how to trigger it with plain requests, and what it needs before you rely on it.
Post-Release Babysitterthedotmack/claude-mem · Apache-2.0 Download the skill Quick start
- 1Download the ZIP. Download the childhood skill ZIP from this page.
- 2Unzip the skill. Unzip the folder so that it contains SKILL.md.
- 3Place it in a skills folder. Copy the folder to ~/.claude/skills/ for personal use or to .claude/skills/ in a project.
- 4Or upload it in the apps. In the Claude apps, go to Settings → Capabilities → Skills and upload the skill.
- 5Confirm prerequisites. Check that you have git access, npm registry access and a local claude-mem plugin installation if you want installed-plugin checks.
- 6Ask after a release. Tell the agent which release shipped and whether you want a specific observation window.
- 7Review the findings. Read the final report for verified findings, pending items and coverage gaps before approving any next action.
What the childhood skill makes the agent do
A successful merge or publish command is only the start of the work. The skill tells the agent to confirm the release identity first (repository, version, tag commit, merge commit, production deployment and affected delivery surfaces), using the release receipt and actual provider or registry state rather than guesses from the current branch.
It then verifies what users actually receive: the exact registry version and its distribution tag, the artifact contents and a small installed CLI smoke test, the live production alias and health of the changed user path, and the plugin registration the user has enabled. It watches CI for the release commit and late issue reports for an observation window, and it verifies any problem before changing code.
- Without a requested window, it watches through 30 minutes after the release first becomes public or production is ready, whichever is later.
- Findings are separated into release regressions, pre-existing issues, operational failures and resolved or transient signals.
- Monitoring does not authorize a new release, rollback, production migration or external announcement. Those still need the user's approval.
Install the childhood skill from the ZIP
Download the ZIP from this page and unzip it. The skill is a folder containing a SKILL.md file. Put it either in your user skills folder, so it is available everywhere, or in a project's own skills folder, so it applies only to that repository.
For a personal install, place the unzipped folder under ~/.claude/skills/. For a project install, place it under the project's .claude/skills/ directory. If you use the Claude apps instead, upload the skill under Settings → Capabilities → Skills.
- Personal: ~/.claude/skills/
- Project: .claude/skills/ inside the repository
- Claude apps: Settings → Capabilities → Skills, then upload
Trigger the skill with natural requests
You do not need a special command. Ask the agent in plain language after a release has merged or been published. The skill's description matches requests about babysitting a shipped release, verifying published packages, production behavior, installed plugins, and late CI or user reports.
Be specific about the release so the agent can identify it. If you want a fixed observation window, say so; otherwise the default window applies. Use the separate babysit skill for pre-release PR readiness, since the description assigns that work there.
- "Babysit the release we just published and tell me when the observation window is clean."
- "Verify the npm package for the new version and check that the production service serves the same commit."
- "Watch CI and new issue reports for the next hour after this deployment."
- "Check that the installed plugin on this machine picks up the new version."
Know what the skill needs before you start
The skill assumes a working release pipeline. Most checks depend on git access to the repository, read access to the npm registry for claude-mem, access to the production hosted service where it applies, and a local claude-mem plugin installation for installed-plugin checks. An optional work_state_write tool can record state if it is available.
If your project has no package registry, no hosted alias and no plugin cache, most of the checks will not apply and the skill will be of little use.
- git and access to the repository
- Package registry access (the npm registry for claude-mem)
- Access to the production hosted service and its authenticated interface, where applicable
- A local claude-mem plugin installation for installed-plugin checks
- Optional: work_state_write tool for recording state
Safety notes: scripts and network
The skill does not run bundled scripts. It does use the network, because its checks query the registry, CI, the production service and user reports. Keep probes read-only unless writes are part of the explicit test scope, and never let the agent print credentials or private result bodies.
The skill is written around claude-mem's tooling and conventions. Review its repair and release steps before letting an agent act on them in your repository, and do not treat the monitoring role as permission for rollbacks or announcements.
- No bundled scripts are run by this skill.
- Network access is used for registry, CI, production and report checks.
- Published versions and existing tags stay immutable; fixes go through the normal PR and release process.
Tips for getting useful results
Give the agent the release version, tag or merge commit up front. The skill is strict about not guessing from the current branch, so the more identity you provide, the less it has to infer.
When a registry or API is unavailable, the skill treats that as a coverage gap rather than proof of health. Expect it to report partial verification with a next action instead of a clean bill of health.
- Name the version, tag or commit you shipped.
- Say whether you want a fixed observation window.
- Expect pending items to be reported as pending, not resolved.
What people use it for
- Verifying that the npm package for a new claude-mem version has the expected files and that a small CLI smoke test runs from an isolated directory.
- Confirming the production hosted service serves the intended commit and that the changed user path still works.
- Checking that an installed claude-mem plugin on a user's machine has picked up the new cache and is not talking to a stale worker.
- Watching CI for late failures on the release commit and triaging new issue reports against the shipped code.
- Keeping a pending webhook or infrastructure failure on record with its owner and next action, instead of retrying it repeatedly.
Frequently asked questions
Is the childhood skill free?
The skill is published under the Apache-2.0 licence, as stated for the claude-mem repository. Check the licence file in the source repository for the full terms before you reuse or modify it.
Do I need claude-mem to use the childhood skill?
The skill is written around the claude-mem project and expects its tooling and conventions. Its checks rely on the claude-mem npm package, its hosted service and its plugin installation, so it is of limited use for unrelated projects.
Where do I put the childhood skill folder?
For a personal install, put it in ~/.claude/skills/. For a single project, put it in that project's .claude/skills/ directory. In the Claude apps, upload it under Settings → Capabilities → Skills.
Does the skill roll back or publish a new release on its own?
No. The skill states that monitoring does not authorize another release, rollback, production migration or external announcement. It prepares and validates a concrete fix before asking for any authorization still missing.
What is the difference between childhood and babysit?
Childhood watches a release after it has merged, published or deployed. The babysit skill handles PR readiness before release, and childhood points to it for repair PRs.
Why does the skill say a check is pending or unverified?
Publication can lag behind a successful command, and some checks may not be exercised. The skill labels those items as pending or unverified rather than marking them healthy, so you know the coverage gap.
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
Use the childhood skill if you maintain claude-mem or a similar plugin and ship packages and hosted services, and you want a structured checklist for the hours after a release. Pick something else if your project has no release pipeline, or if you want an agent to handle pre-release PR readiness, which belongs to the babysit skill.
Download the skill