Core Concepts

ClaudeKit’s power comes from three interconnected systems: Agents, Commands, and Skills.

Agents

Specialized AI assistants with focused expertise.

Available Agents:

  • Planner: Creates implementation plans with phases, tasks, success criteria
  • Researcher: Analyzes codebases, best practices, technical decisions
  • Tester: Writes and runs tests, analyzes failures, fixes issues
  • Debugger: Investigates bugs, traces root causes, proposes fixes
  • Code Reviewer: Reviews code quality, security, performance
  • Docs Manager: Creates and maintains documentation
  • UI/UX Designer: Designs interfaces following aesthetic principles
  • Copywriter: Writes marketing copy, product descriptions, content

How Agents Work:

  1. You invoke a command (e.g., /ck:cook "add feature")
  2. Command spawns relevant agents in sequence or parallel
  3. Agents collaborate via shared context (plans, code, test results)
  4. Output consolidated and presented to you

Skills & Commands

Slash commands and skills that trigger agent workflows.

Note: Skills like /ck:git, /ck:fix, /ck:cook were migrated from commands. Slash syntax unchanged.

Categories:

  • Core: /ck:cook, /ck:plan, /ck:bootstrap, /ck:ask, /ck:scout
  • Fix: /ck:fix (auto/quick/parallel/review), /ck:debug
  • Design: ai-artist skill, ai-multimodal skill, frontend-design skill
  • Git: /ck:git cm, /ck:git cp, /ck:git pr
  • Docs: /ck:docs init, /ck:docs update, /ck:docs summarize
  • Content: Use copywriting skill (quality, fast, CRO workflows)
  • Plan: /ck:plan, /ck:plan --two, /ck:plan --hard
  • Integrate: /integrate:polar, /integrate:sepay
  • Skill: Use skill-creator skill

Example Workflow:

/ck:plan "add payment processing with Stripe"
# → Planner agent creates detailed plan

/ck:cook
# → Reads plan, spawns developer + tester agents, implements feature

/ck:fix
# → Debugger analyzes any test failures, fixes issues

/ck:git cm
# → Git manager stages, commits, pushes changes

Skills

Reusable knowledge modules injected into agent context.

Available Skills:

  • Frontend: Next.js, Tailwind, shadcn/ui
  • Backend: PostgreSQL, Docker
  • AI: Gemini Vision, Canvas Design
  • Auth: Better Auth
  • Ecommerce: Shopify
  • Tools: FFmpeg, ImageMagick, MCP Builder

How Skills Work:

  1. Skill files installed to ~/.claude/skills/ (global) or .claude/skills/ (local project)
  2. Sourced from AgentKit kits — install via ck init -g --kit engineer
  3. Skill provides agent with best practices, examples, gotchas
  4. Agent makes better decisions (uses right patterns, avoids common mistakes)

Creating Custom Skills:

"Create a new skill for FastAPI best practices"
# Use skill-creator skill → Creates new skill with structure, references, examples

How They Work Together

Example: Adding Authentication

You: /ck:cook "add authentication with Better Auth"

System:
1. Detects Better Auth skill → injects into context
2. Spawns Planner agent → creates implementation plan
3. Planner references Better Auth skill → uses correct setup pattern
4. Spawns Developer agent → writes code following skill guidelines
5. Spawns Tester agent → writes tests based on skill examples
6. Output: Fully implemented, tested authentication

CLAUDE.md Configuration

All agent behaviors, skills, and workflows configured in .claude/CLAUDE.md:

# CLAUDE.md
## Agents
- planner: Creates implementation plans
- researcher: Analyzes technical decisions

## Skills
- better-auth: Authentication framework patterns
- nextjs: Next.js App Router best practices

## Workflows
- /ck:plan → /clear → describe task: Plan → cook skill auto-implements
- fix skill: Debug → Fix → Test → Commit

Next Steps