Skip to content

slash-commands.md

Latest commit

 

History

History
384 lines (273 loc) · 10.2 KB

slash-commands.md

File metadata and controls

384 lines (273 loc) · 10.2 KB

Babysitter Slash Commands Reference

Version: 1.0 | Last Updated: 2026-03-05

Complete reference for Babysitter slash commands in Claude Code.

Quick Overview

Babysitter provides two tiers of slash commands:

Core Modes — Four ways to run orchestration, each with different levels of autonomy:

Mode Command Autonomy Best For
Interactive /babysitter:call You approve at breakpoints Learning, critical workflows
YOLO /babysitter:yolo Full auto, no breakpoints Trusted tasks, shipping fast
Forever /babysitter:forever Continuous loop with sleep Monitoring, periodic tasks
Plan /babysitter:plan Planning only, no execution Review before committing

Utility Commands — Setup, diagnostics, and tools:

Command Purpose
/babysitter:user-install Set up your profile and preferences
/babysitter:project-install Onboard a project for babysitting
/babysitter:doctor Diagnose run health and issues
/babysitter:observe Launch real-time monitoring dashboard
/babysitter:assimilate Import external methodologies
/babysitter:help Documentation and guidance

Core Modes

These are the primary ways to invoke Babysitter. Same engine, different behaviors.

/babysitter:call

The default mode. Interactive orchestration with human-in-the-loop approval.

/babysitter:call build a REST API with authentication using TDD

What it does:

  1. Interviews you to understand requirements
  2. Creates a custom process tailored to your request
  3. Asks for confirmation before executing
  4. Pauses at breakpoints for your approval
  5. Iterates until quality targets are met

When to use:

  • First time using Babysitter
  • Critical workflows where you want oversight
  • Learning how processes work
  • Any task where you want to steer decisions

Breakpoint behavior: Pauses and asks you to approve/reject before continuing.

/babysitter:yolo

Ship while you sleep. Full autonomous execution without breakpoints.

/babysitter:yolo add dark mode to the entire frontend

What it does:

  1. Parses your request directly (no interview)
  2. Creates and executes the process
  3. Auto-approves all breakpoints
  4. Iterates until completion or failure

When to use:

  • Tasks you trust Babysitter to handle
  • Overnight or background work
  • When you don't want interruptions
  • After you've validated the approach with /babysitter:plan

Breakpoint behavior: Auto-approves everything. No human interaction required.

The name says it all. YOLO mode is for when you trust the process and want results without babysitting the babysitter.

/babysitter:forever

Set it and forget it. Never-ending orchestration for continuous tasks.

/babysitter:forever monitor support tickets and auto-respond to common questions

What it does:

  1. Creates a process with an infinite loop
  2. Uses ctx.sleep() between iterations
  3. Runs continuously until manually stopped
  4. Perfect for periodic, ongoing work

Example use cases:

  • Monitor and process support tickets every 4 hours
  • Daily code review of new PRs
  • Continuous security scanning
  • Periodic dependency updates
  • Log analysis and alerting

How it works internally:

// Forever mode creates processes like this:
while (true) {
  await processTickets();
  await ctx.sleep({ hours: 4 }); // Wake up in 4 hours
}

To stop a forever run: Close the session or use Ctrl+C.

/babysitter:plan

Look before you leap. Create and review the process without executing it.

/babysitter:plan migrate the database from MySQL to PostgreSQL

What it does:

  1. Interviews you about requirements (same as /call)
  2. Creates the complete process definition
  3. Generates .diagram.md and .process.md visualizations
  4. Stops there — no run is created or executed

When to use:

  • Complex migrations or refactors
  • When you want to review the approach first
  • Team discussions about workflow
  • Understanding what Babysitter would do

After planning:

  • Review the generated process files
  • Modify if needed
  • Run with /babysitter:call when ready

Utility Commands

Setup, diagnostics, and tools to support your workflow.

/babysitter:user-install

First-time setup. Creates your personal profile for better orchestration.

/babysitter:user-install

What it does:

  1. Installs dependencies (SDK, jq, etc.)
  2. Interviews you about your specialties and preferences
  3. Creates ~/.a5c/user-profile.json with:
    • Your expertise areas
    • Breakpoint tolerance (how much oversight you want)
    • Tool preferences
    • Communication style
  4. Configures optimal settings for your workflow

Run this once when you first start using Babysitter. Your profile personalizes every future run — fewer questions, better-matched processes.

/babysitter:project-install

Onboard a project. Set up a codebase for babysitting.

/babysitter:project-install

What it does:

  1. Researches your codebase structure
  2. Interviews you about project goals and workflows
  3. Creates .a5c/project-profile.json with:
    • Project architecture
    • Tech stack
    • Testing frameworks
    • CI/CD configuration
  4. Installs SDK dependencies in .a5c/
  5. Optionally configures CI/CD integration

Run this once per project. The project profile helps Babysitter make better decisions about testing, deployment, and code style.

/babysitter:doctor

Diagnose issues. Comprehensive health check for babysitter runs.

/babysitter:doctor
/babysitter:doctor run-20260125-143012

What it does:

Performs 10 diagnostic checks:

  1. Run Discovery — Finds and validates run metadata
  2. Journal Integrity — Verifies checksums, sequence, timestamps
  3. State Cache Consistency — Checks state matches journal
  4. Effect Status — Identifies stuck or errored tasks
  5. Lock Status — Detects stale or orphaned locks
  6. Session State — Finds active sessions, detects runaway loops
  7. Log Analysis — Scans for errors and warnings
  8. Disk Usage — Reports size, finds oversized files
  9. Process Validation — Verifies entrypoint and SDK dependency
  10. Hook Execution Health — Confirms hooks are running

Output: Detailed report with PASS/WARN/FAIL for each check, plus specific fix commands.

When to use:

  • Run seems stuck or broken
  • After unexpected errors
  • Before resuming an old run
  • When hooks aren't firing

/babysitter:observe

Real-time visibility. Launch a dashboard to watch what Babysitter is doing.

/babysitter:observe

What it does:

Opens a web-based dashboard showing:

  • Active runs and their status
  • Task progress in real-time
  • Journal events as they happen
  • Orchestration state visualization

Built by the community: This tool was created by @yoavmayer as an observability solution for watching babysitter and agent activity. It launches the @yoavmayer/babysitter-observer-dashboard package.

Technical: Runs a local server and opens your browser. Blocking process — runs until stopped.

/babysitter:assimilate

Resistance is futile. Import external methodologies into Babysitter.

/babysitter:assimilate harness codex
/babysitter:assimilate https://github.com/example/cool-methodology

What it does:

Converts external AI coding tools and methodologies into Babysitter process definitions:

  • Harness integration — Generate SDK bindings for other AI agents (Codex, Gemini CLI, etc.)
  • Methodology import — Transform procedural docs into executable processes with skills and agents

This is for advanced users who want to extend Babysitter and contribute back to the community.

Example workflow:

/babysitter:assimilate harness codex

This generates the integration code for OpenAI Codex. Once working, contribute it back so everyone benefits.

Open opportunities — Who's claiming these?

  • OpenAI Codex
  • Google Gemini
  • GitHub Copilot
  • Cursor IDE
  • Windsurf IDE
  • OpenCode

Join the Hall of Fame: a5c.ai/hall-of-fame

Your credit stays there forever. Who's going to be first?

/babysitter:help

Documentation hub. Get help on any command, process, or concept.

/babysitter:help
/babysitter:help command doctor
/babysitter:help process tdd-quality-convergence
/babysitter:help methodology bmad

What it does:

  • No args: Shows all available commands with descriptions
  • With args: Shows detailed documentation for the specific topic

Argument patterns:

  • command <name> — Help on a slash command
  • process <name> — Help on a process definition
  • skill <name> — Help on a skill
  • agent <name> — Help on an agent
  • methodology <name> — Help on a methodology

Mode Selection Guide

Not sure which mode to use? Here's a decision tree:

Start here
    │
    ├─ First time or unfamiliar task?
    │   └─ Use /babysitter:call (interactive mode)
    │
    ├─ Want to review before executing?
    │   └─ Use /babysitter:plan
    │
    ├─ Trusted task, want hands-off?
    │   └─ Use /babysitter:yolo
    │
    ├─ Continuous/periodic task?
    │   └─ Use /babysitter:forever
    │
    └─ Something's broken?
        └─ Use /babysitter:doctor

Common Patterns

Quality-targeted development

/babysitter:call build a user auth system with TDD targeting 90% quality

Overnight autonomous work

/babysitter:yolo refactor the entire codebase to use TypeScript strict mode

Daily automation

/babysitter:forever check for security vulnerabilities in dependencies every 24 hours

Plan-then-execute

/babysitter:plan migrate from Express to Fastify
# Review the generated process...
/babysitter:call proceed with the migration

See Also