# Babysitter

> Orchestrate complex multi-step workflows using babysitter SDK

- Skill: `majiayu000/babysitter` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds add majiayu000/babysitter`
- Raw SKILL.md: https://api.skillmd.com/api/skills/majiayu000/babysitter/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: majiayu000 (https://skillmd.com/u/majiayu000)
- Updated: 2026-09-09
- Page: https://skillmd.com/skills/majiayu000/babysitter

---


# Babysitter Orchestration

Use this skill to orchestrate complex, multi-step coding workflows through babysitter's process-driven execution engine.

## How It Works

1. Babysitter creates a **run** from a **process definition** (JavaScript file with tasks)
2. Each iteration produces **effects** (tasks to execute)
3. You execute the effects and post results back
4. The loop continues until the run completes

## Starting a Run

Use the SDK bridge directly (imported from the extension):
- `createNewRun()` to create a run
- `iterate()` to advance the orchestration
- `postResult()` to report effect outcomes

## Effect Types

| Kind | Action |
|------|--------|
| agent | Execute as sub-agent task |
| node | Run Node.js script |
| shell | Run shell command |
| breakpoint | Ask user for approval |
| sleep | Wait until timestamp |
| skill | Expand and execute skill |

## Completion

When the run completes, the SDK returns a completion proof. Output it wrapped in `<promise>PROOF_VALUE</promise>` tags.