cold-start-generator

Cold Start Generator

Analyzes a codebase and generates everything needed for AI-assisted autonomous development: documentation, skills, templates, and onboarding materials.

When to Use

• Setting up a new project for AI-assisted development
• Onboarding a new developer to an existing codebase
• Making any microservice codebase "AI-development ready"
• When someone says "make this repo work with Claude Code"

Primary Goal

Produce a self-contained cold-start package that allows any developer to clone a repo, point Claude Code at the cold-start doc, and immediately be productive — including autonomous feature development.

What Gets Generated

Output Package:
├── CLAUDE.md                          # Repo-level AI context (architecture, commands, conventions)
├── docs/
│   ├── COLD-START.md                  # Entry point: "Start here"
│   ├── REQUIREMENTS-TEMPLATE.md       # Machine-executable requirements format
│   ├── DEVELOPMENT-PROCESS.md         # The 4-phase process
│   └── SERVICE-MAP.md                 # Service inventory with responsibilities
└── .claude/skills/                    # Custom skills for this project
    ├── requirements-discovery/        # Tailored to this project's patterns
    │   └── SKILL.md
    └── develop-and-test/              # Tailored to this project's test infrastructure
        └── SKILL.md

Analysis Process

Step 1: Codebase Reconnaissance

Spend significant time understanding before generating anything:

1. Identify the architecture pattern:

   • Monolith? Microservices? Serverless? Monorepo?
   • How do services communicate? (HTTP, message bus, events, gRPC)
   • What's the database strategy? (shared DB, per-service DB, gateway)

2. Map the service inventory:

   • List every deployable unit
   • Identify ports, entry points, health checks
   • Map dependencies between services

3. Identify the service skeleton:

   • Read 2-3 services' entry points
   • Extract the common initialization pattern
   • Document: what's the same, what varies