Ralph for GitHub Copilot CLI

An autonomous AI agent loop that runs GitHub Copilot CLI repeatedly until all PRD items are complete. Each iteration is a fresh Copilot instance with clean context. Memory persists via git history, progress.txt, and prd.json.

Based on Geoffrey Huntley’s Ralph pattern and adapted from Ryan Carson’s Amp implementation.

Prerequisites

• GitHub Copilot CLI installed and authenticated (npm install -g @github/copilot)
• Active Copilot subscription (Pro, Pro+, Business, or Enterprise)
• jq installed (apt-get install jq on Linux, brew install jq on macOS)
• A git repository (Ralph commits progress and uses git history for memory)

Quick Start

Copy files to your project, then run:

# In your project directory
cp /path/to/copilot-ralph/ralph.sh ./
cp /path/to/copilot-ralph/prompt.md ./
cp /path/to/copilot-ralph/prd.json.example ./prd.json

# Edit prd.json with your user stories
./ralph.sh

Ralph runs in the current directory and commits to the current git repo.

PRD Format

Create a prd.json file with your user stories:

{
  "projectName": "Landing Page",
  "branchName": "ralph/landing-page",
  "userStories": [
    {
      "id": "US-001",
      "title": "Create hero section",
      "priority": 1,
      "description": "Build hero with headline and CTA",
      "acceptanceCriteria": ["Full viewport height", "CTA scrolls to signup"],
      "passes": false
    }
  ]
}

See prd.json.example for a complete example.

What Ralph Does

./ralph.sh [max_iterations]

Default is 10 iterations. Each iteration:

1. Picks the highest priority story where passes: false
2. Implements that single story
3. Runs quality checks
4. Commits if checks pass
5. Marks story as passes: true
6. Appends learnings to progress.txt
7. Repeats until all stories pass

How It Works

┌─────────────────────────────────────────────────────────────┐
│                                                             │
│   prd.json ─────┬───────── progress.txt ─────── git history │
│                 │                                           │
│                 ▼                                           │
│        copilot -p "$PROMPT" --allow-all-tools               │
│                 │                                           │
│                 ▼                                           │
│        Check for COMPLETE                │
│                 │                                           │
│         ┌──────┴──────┐                                     │
│         ▼             ▼                                     │
│      Complete     Continue                                  │
│      Exit 0       Next iteration                            │
│                                                             │
└─────────────────────────────────────────────────────────────┘

Key Files

Critical Concepts

Fresh Context Each Iteration

Each iteration spawns a new Copilot CLI instance. Memory persists only via:

• Git history
• progress.txt
• prd.json

Small Tasks

Stories should be small enough to complete in one context window.

Good: Add a form component, Create an API endpoint, Add validation

Too big: Build the dashboard, Add authentication, Refactor the API

Stop Condition

When all stories have passes: true, Ralph outputs COMPLETE and exits.

Models

GitHub Copilot CLI uses Claude Sonnet 4.5 by default.

Available models: Claude Sonnet 4.5, Claude Sonnet 4, GPT-5

# Change model via environment variable
export COPILOT_MODEL=gpt-5
./ralph.sh

Debugging

# Check story status
cat prd.json | jq '.userStories[] | {id, title, passes}'

# View learnings
cat progress.txt

# Check commits
git log --oneline -10

Generating PRDs

Use the PRD skill to generate prd.json files through a guided process:

# Ask Copilot to generate a PRD
copilot "Generate a PRD for a user authentication system"

The skill will:

1. Ask 3-5 clarifying questions with A/B/C/D options
2. Generate a properly structured prd.json
3. Create small, implementable user stories with verifiable acceptance criteria

See skills/prd/SKILL.md for the full specification.

Customizing

Edit prompt.md to add:

• Project-specific quality check commands
• Codebase conventions
• Common gotchas for your stack

References

• Geoffrey Huntley’s Ralph
• Ryan Carson’s Amp implementation
• GitHub Copilot CLI
• Copilot CLI Docs