Playbooks

Playbooks are collections of conventions — team standards that guide how AI assistants work in your projects. They enforce patterns like "always use MiniTest" or "follow RESTful routing."

How playbooks work

1. Platform playbooks are published by crewkit for common frameworks (Rails, Next.js, etc.)
2. Your organization subscribes to relevant playbooks
3. Playbooks are assigned to projects (automatically via stack matching or manually)
4. Conventions are injected into agent prompts during sync

Subscribe to a playbook

Browse available playbooks in the dashboard:

1. Go to Playbooks in the sidebar
2. Click Available to see the marketplace
3. Subscribe to playbooks that match your stack

Or via the CLI:

crewkit playbooks list    # See applicable playbooks

Conventions

Each playbook contains conventions — structured rules with:

• Category — What area it covers (testing, architecture, style, security)
• Key — Unique identifier (e.g., use-minitest)
• Value — The actual convention text
• Anti-patterns — What to avoid

Example convention:

{
  "category": "testing",
  "key": "use-minitest",
  "value": "Use MiniTest for all tests. Do not use RSpec.",
  "anti_patterns": ["require 'rspec'", "RSpec.describe"]
}

AI-powered extraction

crewkit can extract conventions from your existing codebase using AI:

1. Go to a playbook in the dashboard
2. Click Extract Conventions
3. Provide a code sample or repository context
4. Review the extracted conventions before saving

This uses Claude to identify patterns and standards already present in your code.

Stack matching

Playbooks can be automatically matched to projects based on detected framework stacks. When crewkit detects that a project uses Rails, it suggests the Rails playbook. The matching considers:

• Package manifests (Gemfile, package.json, Cargo.toml)
• Framework indicators (config/routes.rb, next.config.ts)
• Directory structure

Convention challenges

During a session, if Claude deviates from a convention, the hook system can detect it and log a challenge. This data feeds back into analytics so you can see which conventions are frequently challenged and may need revision.

Next steps

• Experiments
• Hooks
• Team management