---

name: ralph description: This skill should be used when the user invokes "/ralph" to initialize a project for autonomous development. Creates .ralph/ directory with stories.json, progress.txt, and learnings.txt for the Ralph loop. version: 1.1.0

Ralph Skill

Initialize a project for the Ralph autonomous development loop.

CRITICAL: NEVER start implementing within this session. After setup is complete, instruct the user to run ralph from their terminal to begin implementation.

Instructions

When the user invokes /ralph:

Scope Detection

First, analyze the user's request to determine scope:

Single phase indicators (create stories directly):

• One clear feature or component
• Can be broken into 3-5 stories
• No distinct "layers" or "stages"

Multi-phase indicators (offer roadmap):

• Multiple distinct areas (auth, API, frontend, etc.)
• Would need 10+ stories
• Natural phases or stages mentioned
• Words like "full", "complete", "end-to-end"

If multi-phase detected, ask:

This looks like it could be split into phases:
1. [Phase name]
2. [Phase name]
...

Should I create a roadmap with phases, or keep it as one set of stories?

If user chooses roadmap:

1. Create .ralph/ directory if needed
2. Create .ralph/roadmap.json with phase structure (see format below)
3. Create .ralph/learnings.txt if it doesn't exist
4. Do NOT create stories.json or progress.txt yet
5. Update .gitignore
6. Tell user to run ralph --roadmap

If .ralph/ already exists:

Ask the user what they want to do:

1. Add stories - Keep existing stories in stories.json and add new ones. Progress and learnings are preserved.
2. Replace stories - Start fresh with new stories. Then ask:
   • Keep learnings? (yes → preserve learnings.txt, no → reset it)
   • Progress is always reset when replacing stories.

Setup flow (single phase):

1. Understand the project - Read CLAUDE.md and explore the codebase structure
2. Ask clarifying questions - Understand goals, constraints, and priorities
3. Create .ralph/ directory - If it doesn't exist
4. Generate .ralph/stories.json - Create or update the task list with user stories
5. Create .ralph/progress.txt - Initialize (or reset if replacing)
6. Handle .ralph/learnings.txt - Preserve existing, create if missing, or reset if user chose to
7. Update .gitignore - Add .ralph to .gitignore (create if needed)

.ralph/roadmap.json Format

{
  "project": "Project Name",
  "phases": [
    {
      "id": 1,
      "title": "Phase Title",
      "description": [
        "Description bullet 1",
        "Description bullet 2"
      ],
      "complete": false
    },
    {
      "id": 2,
      "title": "Another Phase",
      "description": [
        "Description bullet 1"
      ],
      "complete": false
    }
  ]
}

.ralph/stories.json Format

{
  "project": "Project Name",
  "description": "One-line description",
  "stories": [
    {
      "id": "PROJ-001",
      "title": "Short title",
      "description": "Detailed description of what needs to be built",
      "acceptance_criteria": [
        "Criterion 1",
        "Criterion 2",
        "Tests pass"
      ],
      "priority": 1,
      "passes": false
    }
  ]
}

.ralph/progress.txt Format

# {Project Name} Progress Log

---

.ralph/learnings.txt Format

Only create this file if it doesn't already exist. Preserve existing learnings across ralph runs.

# {Project Name} Learnings

## Codebase Patterns
(Patterns discovered during implementation will be added here)

## Gotchas
(Project-specific warnings will be added here)

---

Guidelines

Story Sizing

Each story must be completable in ONE iteration (one Claude context window).

Right-sized stories:

• Add a database migration and model
• Create a single component
• Add an API endpoint with validation
• Implement a service class

Too big (split these):

• "Build the dashboard" → split into: model, list view, detail view, actions
• "Refactor the API" → split into one story per endpoint

Rule of thumb: If you can't describe the change in 2-3 sentences, split it.

Story Ordering

Stories execute in priority order. Earlier stories must not depend on later ones.

Correct order:

1. Database migrations and models
2. Service classes and business logic
3. Controllers and routes
4. UI components that use the above

Acceptance Criteria

Each criterion must be verifiable, not vague.

Good:

• "User model has email and password fields"
• "API returns 401 for unauthenticated requests"
• "Feature test covers the happy path"

Bad:

• "Works correctly"
• "Good user experience"
• "Handles edge cases"

Always Include

For every story, include at least one of:

• "Tests pass" (for backend logic)
• "Feature test covers the happy path" (for new features)

Project Detection

Detect project type to set appropriate defaults:

Laravel projects (has artisan file):

• Test command: php artisan test --compact
• Story IDs: PROJ-001 format (use project name prefix)

Node.js projects (has package.json):

• Test command: check scripts.test in package.json
• Story IDs: PROJ-001 format

Python projects (has pyproject.toml or setup.py):

• Test command: pytest
• Story IDs: PROJ-001 format

Phase Mode (called from ralph --roadmap)

When invoked with phase context, you'll receive a prompt like:

Run /ralph for this phase:

Phase: Phase 1: Data Models
  - User and Project models with migrations
  - Basic validation and relationships

Create stories only for this phase. Preserve learnings.txt if it exists.

In this mode:

• Create stories ONLY for this phase's scope
• Always reset .ralph/progress.txt (new phase = fresh progress)
• Preserve .ralph/learnings.txt if it exists
• Do NOT ask clarifying questions - use the phase description as-is
• After setup, output: <ready>PHASE_READY</ready>

After Running

Tell the user what happened:

If roadmap created:

• .ralph/roadmap.json - created with N phases
• .ralph/learnings.txt - created (if new)
• .ralph added to .gitignore
• Run ralph --roadmap from terminal to start

If stories created:

• .ralph/stories.json - created/updated with N stories (N new if adding)
• .ralph/progress.txt - initialized or reset
• .ralph/learnings.txt - preserved, created, or reset (based on user choice)
• .ralph added to .gitignore (if not already)
• Run ralph from terminal to start the autonomous loop
• Run ralph 1 for a single iteration

Do NOT start implementing. The user must run ralph from their terminal.