Agentic AI Atlasby a5c.ai
OverviewWikiGraphFor AgentsEdgesSearchWorkspace
/
GitHubDocsDiscord
iiRecord
Agentic AI Atlas · Quickstart: Your First Babysitter Run
page:docs-user-guide-getting-started-quickstarta5c.ai
Search record views/
Record · tabs

Available views

II.Record viewspp. 1 - 1
overviewarticlejsongraph
III.Related pagespp. 1 - 1
II.
Page reference

page:docs-user-guide-getting-started-quickstart

Reading · 9 min

Quickstart: Your First Babysitter Run reference

Docs(../index.md) › Getting Started(./README.md) › Quickstart

Pagewiki/docs/user-guide/getting-started/quickstart.mdOutgoing · 0Incoming · 1

Docs › Getting Started › Quickstart

Quickstart: Your First Babysitter Run

**Time:** ~10 minutes | **Level:** Beginner | **Prerequisites:** Installation complete

**End state:** in about 10 minutes you will have run a deterministic Babysitter process once — where the orchestrator only did what the code permitted, stopping after every step before deciding what's next — and seen the result land on disk, all from a single command in your harness.

Welcome! In this quickstart, you will build a calculator module using Test-Driven Development (TDD) with Babysitter. By the end, you will have experienced:

  • **Deterministic, obedient execution** - the orchestrator only ran what the process permitted, with a mandatory stop after each step (enforcement, not assistance)
  • **Journal-based persistence** - everything recorded in an immutable, replayable journal
  • **A quality gate in action** - quality convergence (iterate until the target is met) as one of the gates the process enforced
  • The TDD workflow (tests first, then implementation)

**Note:** TDD Quality Convergence is the full name; we use "TDD" as shorthand throughout this guide.

On this page

  • What You Will Build
  • Before You Begin
  • Step 1: Launch Your First Run
  • Step 2: Watch the Magic Happen
  • Step 4: Explore the Journal
  • Bonus: Experience Session Resumption
  • Common First-Run Issues

Pick your harness first

This quickstart uses Claude Code's /babysitter:call token. **The command is the same on every harness — only the entry token differs** ($babysitter:call on Codex, $call on Cursor/Copilot, /call on the thin-alias harnesses, /status on opencode). If you are not on Claude Code, substitute your harness's token — see Slash Commands and Modes and the Install Matrix.

Let's get started!

---

What You Will Build

A simple calculator module with:

  • add(a, b) - Add two numbers
  • subtract(a, b) - Subtract two numbers
  • multiply(a, b) - Multiply two numbers
  • divide(a, b) - Divide two numbers (with error handling)

The result will include:

  • Working implementation
  • Test suite with multiple test cases
  • 80%+ quality score achieved through automatic iteration

---

Before You Begin

Step 1: Set Up Your Profile (First Time Only)

If you haven't already, configure your personal preferences:

bash
/babysitter:user-install

This personalizes Babysitter for your workflow - breakpoint frequency, communication style, and expertise areas.

Step 2: Set Up Your Project

In your project directory:

bash
/babysitter:project-install

This analyzes your codebase and configures project-specific settings.

Step 3: Verify Installation

Quick check that everything is working:

bash
# In your terminal
npm list -g @a5c-ai/babysitter
adapters version

# Or run diagnostics
/babysitter:doctor

You should see version numbers (e.g. @a5c-ai/babysitter@6.0.0 and 6.0.0). If not, revisit the installation guide.

Open Your Project

Navigate to your project directory (or create a new one):

bash
# Create a new project directory
mkdir my-babysitter-project
cd my-babysitter-project

# Initialize npm (optional but recommended)
npm init -y

---

Step 1: Launch Your First Run

Open Claude Code in your project directory and enter this command:

Code
/babysitter:call create a calculator module with add, subtract, multiply, and divide functions using TDD with 80% quality target

**Alternative (natural language):**

Code
Use the babysitter skill to build a calculator module with TDD and 80% quality target

What You Should See

Babysitter will start and show output like:

Code
Creating new babysitter run: calculator-20260125-143012
Process: TDD Quality Convergence
Target Quality: 80%

Run ID: 01KFFTSF8TK8C9GT3YM9QYQ6WG
Run Directory: .a5c/runs/01KFFTSF8TK8C9GT3YM9QYQ6WG/

Babysitter is now orchestrating your TDD workflow!

---

Step 2: Watch the Magic Happen

Sit back and observe as Babysitter works through the TDD methodology:

Phase 1: Research (~30 seconds)

Code
[Phase 1] Research
- Analyzing project structure... done
- Checking existing patterns... done
- Identifying test framework... done

Babysitter examines your codebase to understand the context.

Phase 2: Specifications (~1 minute)

Code
[Phase 2] Specifications
- Defining calculator interface...
- Specifying test cases...
- Creating implementation plan...

Specifications complete:
- 4 functions defined
- 12 test cases planned
- Jest test framework selected

Babysitter creates a clear specification before coding.

Phase 3: TDD Implementation Loop

This is where the magic happens. Babysitter iterates until quality is achieved:

Iteration 1:

Code
[Iteration 1/5] Starting TDD implementation...

Writing tests:
- add.test.js: 3 test cases
- subtract.test.js: 3 test cases
- multiply.test.js: 3 test cases
- divide.test.js: 3 test cases (including error handling)

Implementing code:
- calculator.js: add, subtract, multiply, divide functions

Quality checks:
- Tests: 11/12 passing
- Coverage: 75%
- Linting: 2 warnings

Quality Score: 72/100 (target: 80)
Below target, continuing...

Iteration 2:

Code
[Iteration 2/5] Refining implementation...

Fixes:
- Fixed divide by zero test
- Improved edge case handling
- Resolved lint warnings

Quality checks:
- Tests: 12/12 passing
- Coverage: 92%
- Linting: 0 warnings

Quality Score: 88/100 (target: 80)
Target achieved!

---

Step 3: Review the Results

When Babysitter completes, you'll see a summary:

Code
Run completed successfully!

Summary:
- Iterations: 2 of 5
- Final Quality Score: 88/100
- Test Coverage: 92%
- Tests: 12 passing
- Duration: 3m 45s

Files created:
- calculator.js
- calculator.test.js

Run ID: 01KFFTSF8TK8C9GT3YM9QYQ6WG
Journal: .a5c/runs/01KFFTSF8TK8C9GT3YM9QYQ6WG/journal/*.json

Explore What Was Created

Check your project directory:

bash
ls -la

You should see new files:

Code
calculator.js       # Your calculator implementation
calculator.test.js  # Test suite
.a5c/              # Babysitter run data

View the Calculator Code

Open calculator.js:

javascript
// calculator.js - Created by Babysitter TDD workflow

function add(a, b) {
  return a + b;
}

function subtract(a, b) {
  return a - b;
}

function multiply(a, b) {
  return a * b;
}

function divide(a, b) {
  if (b === 0) {
    throw new Error('Cannot divide by zero');
  }
  return a / b;
}

module.exports = { add, subtract, multiply, divide };

Run the Tests Yourself

bash
npm test
# or
npx jest

**Expected output:**

Code
PASS  ./calculator.test.js
  Calculator
    add
      ✓ adds two positive numbers
      ✓ adds negative numbers
      ✓ adds zero
    subtract
      ✓ subtracts two numbers
      ...

Test Suites: 1 passed, 1 total
Tests:       12 passed, 12 total

---

Step 4: Explore the Journal

Every action Babysitter took is recorded in the journal. Let's explore:

bash
# View the journal files
ls .a5c/runs/01KFFTSF8TK8C9GT3YM9QYQ6WG/journal/*.json

**Sample events from journal JSON files:**

json
{"type":"RUN_CREATED","recordedAt":"2026-01-25T14:30:12Z","data":{"runId":"01KFFTSF8TK8C9GT3YM9QYQ6WG"},"checksum":"a1b2c3"}
{"type":"EFFECT_REQUESTED","recordedAt":"2026-01-25T14:30:13Z","data":{"effectId":"research-001","effectType":"agent"},"checksum":"d4e5f6"}
{"type":"EFFECT_RESOLVED","recordedAt":"2026-01-25T14:30:38Z","data":{"effectId":"research-001","duration":25000},"checksum":"g7h8i9"}
{"type":"EFFECT_REQUESTED","recordedAt":"2026-01-25T14:31:00Z","data":{"effectId":"tdd-impl-001","effectType":"agent","iteration":1},"checksum":"j0k1l2"}
{"type":"EFFECT_RESOLVED","recordedAt":"2026-01-25T14:33:00Z","data":{"effectId":"tdd-impl-001","iteration":1},"checksum":"m3n4o5"}
{"type":"EFFECT_REQUESTED","recordedAt":"2026-01-25T14:33:01Z","data":{"effectId":"tdd-impl-002","effectType":"agent","iteration":2},"checksum":"p6q7r8"}
{"type":"EFFECT_RESOLVED","recordedAt":"2026-01-25T14:34:30Z","data":{"effectId":"tdd-impl-002","iteration":2},"checksum":"s9t0u1"}
{"type":"RUN_COMPLETED","recordedAt":"2026-01-25T14:34:45Z","data":{"status":"success"},"checksum":"v2w3x4"}

This is the audit trail. Every effect request, every resolution - all recorded. Core SDK event types include RUN_CREATED, EFFECT_REQUESTED, EFFECT_RESOLVED, RUN_COMPLETED, RUN_HALTED, and RUN_FAILED. RUN_HALTED means a process intentionally stopped early via ctx.halt(...) and does not receive a completion proof.

---

Step 5: Try a Quick Modification

Let's see how easy it is to extend your calculator. Ask Babysitter to add more features:

Code
/babysitter:call add a power function and square root function to the calculator with TDD

Babysitter will: 1. Analyze the existing calculator 2. Write new tests for power and sqrt 3. Implement the new functions 4. Iterate until quality is achieved

---

What Just Happened?

Let's recap what Babysitter did for you:

Without Babysitter (Manual Approach)

1. You: "Claude, write tests for a calculator" 2. You: "Now implement the calculator" 3. You: "Run the tests... 2 failed. Fix them." 4. You: "Check coverage... too low. Add more tests." 5. You: "Run tests again... passed!" 6. You: (repeat if you want higher quality)

**Time:** 20-30 minutes with multiple back-and-forth interactions

With Babysitter (Automated Approach)

1. You: "/babysitter:call create calculator with TDD, 80% quality" 2. (Babysitter handles everything automatically) 3. Done!

**Time:** ~5 minutes, hands-free

Key Takeaways

1. **Deterministic, Obedient Execution:** The orchestrator only did what your process permitted, stopping after each step to decide what's next — enforcement, not assistance 2. **Complete Audit Trail:** Every action logged in the immutable journal 3. **No Context Loss:** If interrupted, you can replay the journal and resume exactly where you left off 4. **TDD Methodology:** Tests were written before implementation 5. **Quality Convergence (one gate type):** You set an 80% target and the quality gate iterated until it achieved 88%

---

Bonus: Try Different Modes

You just used /babysitter:call — the default interactive mode. Babysitter has four modes, each with different levels of autonomy:

ModeCommandWhen to Use
**Interactive**/babysitter:callWhat you just used. Pauses for approval.
**YOLO**/babysitter:yoloFull auto. Ship while you sleep.
**Forever**/babysitter:foreverNever-ending loops for monitoring tasks.
**Plan**/babysitter:planReview the process before executing.

**Try YOLO mode** for a trusted task:

Code
/babysitter:yolo add input validation to all form fields

No breakpoints, no questions. Babysitter handles everything autonomously.

**Full reference:** Slash Commands Reference

---

Bonus: Experience Session Resumption

One of Babysitter's superpowers is persistence. Let's try it:

Start a Long-Running Task

Code
/babysitter:call build a REST API for task management with authentication, using TDD with 85% quality target and max 10 iterations

Interrupt It

Close Claude Code or press Ctrl+C while it's running.

Resume Later

Open Claude Code again and run:

Code
/babysitter:call resume the babysitter run

or

Code
/babysitter:call resume

Babysitter will: 1. Find the interrupted run 2. Replay the journal to restore state 3. Continue from exactly where it stopped

No work lost!

---

Common First-Run Issues

"Nothing happens after I type the command"

**Cause:** Plugin may not be loaded.

**Solution:** 1. Check /skills shows "babysit" 2. Restart Claude Code if needed 3. Verify plugin is enabled: claude plugin list

"Breakpoint timeout" error

**Cause:** You may have missed the question in the chat or the session timed out.

**Solution:**

  • Scroll up to find the breakpoint question and respond
  • Or resume the run: claude "/babysitter:call resume the babysitter run"

Quality score not reaching target

**Cause:** Target may be too high for the task complexity.

**Solution:**

  • Lower the target (try 70% instead of 90%)
  • Increase max iterations: --max-iterations 10
  • Be more specific in your request

Run seems stuck

**Cause:** Waiting for breakpoint approval.

**Solution:**

  • Look for a question from Claude in your chat
  • Respond to approve and continue the workflow

---

Explore Further

Congratulations! You've completed your first Babysitter run. Here's what to explore next:

Try different prompts

  • /babysitter:call refactor the calculator for better error handling
  • /babysitter:call add comprehensive documentation to the calculator
  • /babysitter:call increase test coverage to 95%

This Week

- TDD (Test-Driven Development) - what you just used - GSD (Get Stuff Done) - faster, less formal - Spec-Kit - specification-driven development

  • **Explore methodologies:**
  • **Configure breakpoints** for approval workflows

Advanced Topics

  • **Custom quality targets** and scoring criteria
  • **Parallel execution** for faster runs
  • **Custom process definitions** (for power users)

---

Quick Reference

Commands used in this quickstart:

bash
# Start a TDD run with quality target (in Claude Code)
/babysitter:call <description> with TDD and <X>% quality target

# Resume an interrupted run
/babysitter:call resume

# View run journal files
ls .a5c/runs/<runId>/journal/*.json

# List all runs
ls .a5c/runs/

---

Summary

In just 10 minutes, you:

  • Ran a deterministic process where the orchestrator only did what your code permitted
  • Built a calculator module with TDD methodology
  • Explored the event journal (complete, replayable audit trail)
  • Learned how to resume interrupted sessions
  • Saw a quality gate converge to its target (one gate type among several)

**Babysitter enforces obedience on agentic work: it turns complex AI workflows into single commands with deterministic, resumable execution.**

Ready to go deeper? Continue to First Run Deep Dive to understand exactly what happened under the hood.

---

Next steps

  • **Next:** First Run Deep Dive — understand exactly what happened under the hood
  • **Previous:** Installation
  • **Related:** Slash Commands — every mode and command token

Article source

The article body is owned directly by this record.

Related pages

No related wiki pages for this record.

Shortcuts

Open overview
Open JSON
Open graph