Claude Code Guide

Skills and guides for autonomous AI-assisted development with Claude Code and Ralph loops.

What This Is

Three skills you can add to any project:

• prd - Create PRDs that Claude can execute autonomously
• setup-claude - Initialize or audit repositories for optimal Claude Code usage
• agent-browser - Headless browser automation and E2E testing

Plus guides on how I actually use Claude Code.

Quick Start

Add Skills to Your Project

# Add the PRD skill
npx cc-guide add-skill prd

# Add the setup/audit skill
npx cc-guide add-skill setup-claude

# Add the browser automation skill
npx cc-guide add-skill agent-browser

# Or add all skills
npx cc-guide add-skill --all

This copies the skills to your project's .claude/skills/ folder.

Use the Skills

In Claude Code:

/prd                    # Create a PRD for a new feature
/setup-claude init      # Set up a new repository
/setup-claude audit     # Audit an existing repository

The Idea

Traditional: You code, AI helps, you iterate forever.

This way: You define the work, AI executes, you review.

Traditional:          Ralph Loop:
Human → AI → Human    Human → PRD → AI Loop → Human Review
     ↑      ↓              ↓           ↓
     └──────┘          (autonomous)  (complete)

The quality of your PRD determines the quality of your output.

What You Need

| Tool | Install | Why | |------|---------|-----| | Claude Code | claude.ai/claude-code | Core CLI | | GitHub CLI | brew install gh | Git operations | | tmux | brew install tmux | Persistent sessions | | Ralph TUI | npm install -g ralph-tui | Task orchestration | | Agent Browser | npm install -g agent-browser | Browser automation |

Guides

• My Claude Workflow - How I actually use Claude Code
• Ralph Loops Explained - Running autonomous development loops
• Best Practices - Lessons learned

Skills

PRD Skill (/prd)

Creates self-verifying PRDs through an interview process:

1. Identifies what type of task you're doing
2. Extracts everything you know via brain dump
3. Asks clarifying questions
4. Confirms understanding
5. Generates PRD.md and prd.json

The output is a PRD that Ralph TUI can execute autonomously.

Agent Browser Skill (/agent-browser)

Headless browser automation for AI agents:

• Navigate, click, fill forms
• Take screenshots and snapshots
• E2E testing workflows
• Web scraping

Essential for any task involving UI verification.

Setup Claude Skill (/setup-claude)

Two modes:

Init - For new repositories (13 phases):

• Environment & tools check (including tmux)
• CLI discovery & authentication
• MCP configuration (context-window aware)
• Skills, subagents, rules, hooks setup
• CLAUDE.md generation
• Ralph TUI configuration

Audit - For existing repositories (8 phases):

• Environment analysis
• CLI and context window audit
• Gap analysis
• Recommendations with interactive fixes

Repository Structure

cc-guide/
├── skills/
│   ├── prd/                  # PRD creation skill
│   │   ├── SKILL.md          # Entry point
│   │   ├── AGENTS.md         # Comprehensive reference
│   │   ├── interview/        # Interview process guides
│   │   └── categories/       # Task-type specific guidance
│   │
│   ├── setup-claude/         # Setup/audit skill
│   │   ├── SKILL.md          # Entry point
│   │   ├── workflows/        # Init & audit workflows
│   │   ├── components/       # Detailed component guides
│   │   ├── reference/        # Reference material
│   │   └── templates/        # Config templates
│   │
│   └── agent-browser/        # Browser automation skill
│       └── SKILL.md          # CLI reference
│
├── guides/
│   ├── my-claude-workflow.md
│   ├── ralph-loops-explained.md
│   └── best-practices.md
│
├── templates/
│   └── prompt.hbs            # Ralph TUI prompt template
│
└── bin/
    └── add-skill.js          # npx cc-guide add-skill

Key Concepts

CLI First, MCP Last

CLIs don't consume context window. MCPs do.

Before enabling any MCP:

• Check if a CLI exists (gh, vercel, supabase, stripe)
• Install and authenticate the CLI
• Only use MCP when there's no CLI alternative

tmux for Persistence

Ralph loops run for hours. tmux means:

• Sessions persist through SSH disconnects
• You can detach and come back later
• Multiple loops can run in parallel

Worktrees for Parallelism

Can't run two Ralph loops on the same codebase. Git worktrees fix this:

git worktree add ../project-feature-a feature-a

Each worktree can run its own tmux session with its own Ralph loop.

License

MIT - See LICENSE

The goal isn't to remove humans from development. It's to elevate humans from typists to architects.