Ralph Init

Initialize a project with the Ralph workflow structure for autonomous AI-driven development.

Ralph Philosophy

Ralph is a loop-based autonomous development workflow. Key principles:

Three Phases:

1. Define - Write specs for each topic of concern (one spec = one topic)
2. Plan - Gap analysis: compare specs vs code, generate prioritized task list
3. Build - Implement one task per loop iteration, commit, repeat

Context is Everything:

• Each loop iteration starts with fresh context
• Use subagents for exploration (preserves main context)
• IMPLEMENTATION_PLAN.md persists state between iterations
• AGENTS.md captures operational learnings

Backpressure Steers Quality:

• Tests, typechecks, lints reject invalid work
• Loop continues until backpressure passes
• "Run tests" in prompt, specific commands in AGENTS.md

Let Ralph Ralph:

• Trust the loop to self-correct through iteration
• Plan is disposable - regenerate when wrong
• Observe failures, add guardrails to prompts/AGENTS.md

Reference: Ralph Playbook

What Gets Created

project/
├── specs/
│   ├── README.md               # Spec writing guide
│   └── (example-spec.md)       # Optional example spec
├── AGENTS.md                   # Project guide (commands, architecture, patterns)
├── IMPLEMENTATION_PLAN.md      # Task list (generated during planning)
├── PROMPT_plan.md              # Planning mode prompt
├── PROMPT_build.md             # Building mode prompt
└── loop.sh                     # Ralph loop script (executable)

Note: AGENTS.md serves as the project's CLAUDE.md - they are the same concept. Some projects symlink CLAUDE.md → AGENTS.md.

Step 1: Gather Project Info

Ask the user:

1. Project name - Name for the header
2. Project description - One-line description
3. Project goal - What this project achieves (for PROMPT_plan.md)
4. Source location - Where is source code? (default: src)
5. Shared utilities - Where are shared components? (default: src/lib)
6. Package manager - bun (default), pnpm, or npm

Step 2: Check Existing Files

Check if target files exist. If so, ask whether to skip or overwrite each.

Step 3: Copy Templates

Templates location: ~/.claude/skills/ralph-init/templates/

Step 4: Replace Placeholders

In AGENTS.md:

• {{PROJECT_NAME}} → project name
• {{PROJECT_DESCRIPTION}} → project description
• {{SOURCE_DIR}} → source location

In PROMPT_plan.md and PROMPT_build.md:

• {{SOURCE_DIR}} → source location (e.g., src)
• {{SHARED_UTILS}} → shared utilities (e.g., src/lib)
• {{PROJECT_GOAL}} → project goal

Step 5: Customize AGENTS.md

After copying, help the user customize:

1. Commands - Update for their package manager choice
2. Architecture - Explore their project structure and fill in the ASCII tree
3. Reference Implementations - Identify good examples in their codebase
4. Database commands - Adjust for their ORM/database

Step 6: Make loop.sh Executable

chmod +x loop.sh

Step 7: Optional CLAUDE.md Symlink

Ask if they want to symlink CLAUDE.md to AGENTS.md:

ln -s AGENTS.md CLAUDE.md

This makes the file discoverable as both names.

Step 8: Summary

Ralph workflow initialized!

Created:
  specs/README.md           - Spec writing guide
  specs/example-auth.md     - Example spec (if requested)
  AGENTS.md                 - Project guide
  IMPLEMENTATION_PLAN.md    - Task list
  PROMPT_plan.md            - Planning prompt
  PROMPT_build.md           - Building prompt
  loop.sh                   - Loop script

Next steps:
  1. Review and customize AGENTS.md for your project
  2. Write specs in specs/ for your features
  3. Run ./loop.sh plan to generate implementation plan
  4. Review the plan
  5. Run ./loop.sh to start building

For interactive development:
  - superpowers:test-driven-development for TDD
  - ultra-frontend for UI components

Templates Overview

AGENTS.md (Comprehensive)

Based on Geoff Huntley's loom AGENTS.md structure:

• Tech stack
• Development philosophy
• Specifications guidance
• Reference implementations
• Commands (dev, validation, database)
• Deployment
• Architecture
• Svelte 5 patterns
• Code quality guidelines
• Testing philosophy
• Common pitfalls
• Troubleshooting
• Operational notes
• Development workflow

example-spec.md

A complete example spec (User Authentication) showing:

• Purpose
• Requirements
• Acceptance criteria
• Edge cases
• API endpoints
• Dependencies

Use this as a reference for writing specs.