Search skills/

Babysitter Claude skills

An npm-installed orchestration runtime that makes coding agents follow workflows defined in JS code, with quality gates, approval breakpoints and a run journal.

a5c-ai/babysitter 2118 skills Planning, TDD & debuggingSkill & agent tooling
Our verdict Recommended with caveats

A real, actively developed runtime with CI and evals, but setup takes several npm packages, the skill library is sprawling and thin per skill, and the maintainers' internal skills ship alongside the user-facing ones.

4/5 Depth expertise shipped ! Security Broad access 43d Maintenance Updated 1 mo ago 1.9K Adoption stars OK License MIT, commercial OK

What Babysitter does

Babysitter is an orchestration runtime, not a single skill. You define a workflow as a JavaScript process, with tasks built using defineTask from @a5c-ai/babysitter-sdk, and Babysitter makes the agent follow it step by step. Quality gates must pass before the run moves on, breakpoints stop for human approval, and every decision is recorded in an immutable journal. Inside a harness you start runs with commands such as /babysitter:call, or /babysitter:yolo for runs without breakpoints. The README says the same processes run on 12 coding harnesses through its Adapters runtime. Claude Code and Codex have dedicated guides, and an internal harness built on the SDK's Pi engine runs in CI without an external agent.

The repo is very large: about 2,100 SKILL.md files and nearly 18,000 files in total. Most skills live under library/methodologies (automaker, bmad-method, cc10x and others). They are short, around 1 KB each with one reference file, and describe steps that processes call. The .claude/skills folder holds the maintainers' own tooling, such as working through GitHub issues assigned to a5c-agent or cataloging public repos that use the SDK. The SDK also includes a token compression layer that the README says cuts context use by 50 to 67 percent.

Good for

  • Teams that want agent workflows defined in code with enforced steps
  • Long multi-step runs that need approval checkpoints and an audit trail
  • Running the same process across several coding harnesses or in CI

Not for

  • People looking for a drop-in prompt skill with no install
  • Small tasks where writing a JS process is more work than the task
  • Environments that cannot run global npm packages or Node 22.13+

What's inside depth 4/5

An SDK, CLIs and 81 eval files back the runtime, but the roughly 2,100 skills average 3.7 KB and most library skills are about 1 KB with a single reference.

Skills
2118
Reference docs
187
Scripts
0
Tests / evals
81
Templates
0
Avg SKILL.md
3.6 KB

Counted from the repository file tree. How we score depth.

Dependencies and costs

  • Node.js 20+ (22.13+ for the host-side adapters CLI)
  • Global npm packages: @a5c-ai/babysitter, optionally @a5c-ai/adapters-cli and @a5c-ai/genty-platform
  • A supported AI coding harness such as Claude Code or Codex
  • Authenticated gh CLI for the issue and catalog skills

Security review

Broad access

This skill asks for more access than most, or sends data to another service. That is usually a design choice rather than bad intent, but you should know before you install it.

  • Pre-approves broad shell commandsFrontmatter allowed-tools pre-approves unrestricted Bash. The same applies to all eight cc10x skills (code-generation, code-review-patterns, debugging-patterns, planning-patterns, session-memory, test-driven-development, verification-before-completion).library/methodologies/cc10x/skills/architecture-patterns/SKILL.md:4
  • Pushes or merges without askingTells the agent to commit and push fixes directly to the staging branch via /babysitter:yolo, with no user interaction or breakpoints..claude/skills/fix-failing-pipelines/SKILL.md:55
  • Pushes or merges without askingProcesses every assigned issue without user selection or breakpoints, pushing branches and opening pull requests automatically..claude/skills/babysit-babysitter-issues/SKILL.md:34

This large monorepo was only partly reviewed: we had its SKILL.md files but none of its code, and its orchestration depends on the external @a5c-ai/babysitter-sdk CLI, whose output the Codex skill follows as instructions. Several bundled cc10x skills pre-approve unrestricted Bash. Two maintenance skills tell the agent to push commits (directly to staging, or to new branches with PRs) without any user interaction.

Worth knowing

  • Runs remote code at an unpinned versionRuns the babysitter CLI via npm exec --yes --package @a5c-ai/babysitter-sdk@$SDK_VERSION, which falls back to 'latest' if versions.json cannot be read..codex/skills/babysit/SKILL.md:40
  • Runs remote code at an unpinned versionREADME instructs running npx --yes @a5c-ai/babysitter-codex without a pinned version.README.md:390
  • Relies on a CLI or package we could not reviewOrchestration instructions come from the output of babysitter instructions:babysit-skill, a CLI whose source is not in the reviewed files..codex/skills/babysit/SKILL.md:55
  • Installs packages or toolsMay globally reinstall @a5c-ai/babysitter-sdk with npm and tells the agent to install jq if it is missing..codex/skills/babysit/SKILL.md:44
  • Makes network requestsQueries the GitHub search API via gh and fetches clawhub.ai pages with curl or a browser to discover repositories..claude/skills/assimilate-popular-workflows/SKILL.md:70
  • Makes network requestsUses authenticated gh code search and gh api calls against GitHub..claude/skills/catalog-babysitter-users/SKILL.md:22
  • Contains compiled binariesThe repo contains a compiled Python bytecode file.packages/atlas/graph/tools/validator/__pycache__/validate.cpython-311.pyc

What we checked

The repository (17885 files) on Oct 29, 2026. Not a full review: 17500 of 17885 files were not downloaded (large repository; we reviewed the skill folders); it contains compiled binaries or archives we cannot read; it relies on code outside the repository: About 17,500 readable repo files that were not provided, including .a5c/orchestrator_scripts/*.py, .a5c/processes/*.mjs, the packages/ sources and the library process .js files; npm package @a5c-ai/babysitter-sdk (babysitter CLI, run via npm exec or global install); npm package @a5c-ai/babysitter-codex (run via npx --yes in the README).

NVIDIA SkillSpector static scan: 11 pattern hits (MCP Rug Pull 3, Data Exfiltration 3, Rogue Agent 2, Agent Snooping 2, Tool Misuse 1). The 'rm -g' hit is an npm uninstall/reinstall repair step for the babysitter CLI. The self-modification and skill-enumeration hits are documentation about editing memory files and paths to SKILL.md files. The star-history URL is a README badge image, and the vitest/playwright npx hits are ordinary test commands, not MCP servers.

A review, not a certification. Skills change after we check them — read the files before installing.

License

MIT, commercial use allowed.

Can it run unattended?

Yes. Runs follow a coded process with quality gates, a journal and a resume command, and yolo mode drops breakpoints, so long unattended runs are the intended use once you have written the process.

Alternatives to Babysitter

Superpowers

obra/superpowers

A 15-skill development methodology that makes coding agents clarify the goal, write a plan, work through it with subagents and verify before calling it done.

Recommended Check first Depth 3/5

Compound Engineering

EveryInc/compound-engineering-plugin

Every's plugin of 36 skills for a brainstorm, plan, build, review and capture-the-learning loop, with an optional hands-off /lfg pipeline.

Recommended Broad access Depth 4/5

Planning with Files

OthmanAdi/planning-with-files

Keeps task_plan.md, findings.md and progress.md on disk and re-injects them via hooks, so a plan survives /clear, compaction and crashes.

Recommended Broad access Depth 4/5

wshobson/agents

wshobson/agents

A multi-harness plugin marketplace with 94 plugins, 202 agents, 185 skills and 105 commands for Claude Code, Codex, Cursor, OpenCode, Copilot, Antigravity and Pi.

With caveats Check first Depth 2/5

FAQ

What is Babysitter for Claude Code?

Babysitter is an open-source orchestration runtime from a5c-ai. You write a workflow as a JavaScript process, and it makes the agent follow each step, pass quality gates, stop at approval breakpoints and log decisions in a journal. It ships as an npm CLI plus harness plugins for Claude Code, Codex, Cursor and others.

How do I install Babysitter?

Install the main CLI with npm install -g @a5c-ai/babysitter on Node 20 or newer, then add the plugin for your harness following the install matrix in the docs. If you want to run harnesses from your shell, also install @a5c-ai/adapters-cli, which needs Node 22.13 or newer.

Is Babysitter safe to install?

We found no hidden network calls or credential reads in the scanned files. The README uses an unpinned npx --yes command, and yolo mode runs without breakpoints, so review a process before letting it run unattended. One bundled maintainer skill pushes branches and opens PRs automatically, but only against the Babysitter repo.

Babysitter vs Superpowers: which should I use?

Superpowers is a set of markdown skills that shape how the agent plans, tests and reviews, with nothing extra to install. Babysitter is a runtime where you define the process in code and the CLI enforces it. Pick Superpowers for a lighter process layer, Babysitter if you want coded, repeatable workflows with gates and a journal.

Sources: GitHub repository (stars, license, file tree, last push Sep 16, 2026). Review based on the README and SKILL.md files as published.