Claude Kit

A comprehensive toolkit for Claude Code to accelerate development workflows for teams working with Python and JavaScript/TypeScript.

Features

• 20 Specialized Agents - From planning to deployment
• 43 Skills - Auto-triggered by context: framework, language, methodology, patterns, workflows, optimization (all with YAML frontmatter and bundled resources)
• 7 Behavioral Modes - Task-specific response optimization
• Token Optimization - 30-70% cost savings with compressed output modes
• MCP Integrations - Context7, Sequential Thinking, Playwright, Memory, Filesystem
• Context Management - Project indexing, checkpoints, parallel tasks

Quick Start

1. Copy the .claude folder to your project root
2. Customize .claude/CLAUDE.md for your project
3. Start working — skills auto-trigger based on what you ask ("implement X", "fix Y", "review Z")

Directory Structure

.claude/
├── CLAUDE.md              # Project context (customize this!)
├── settings.json          # Hooks, permissions, and MCP config
├── agents/                # 20 specialized agents
├── modes/                 # 7 behavioral mode definitions
├── mcp/                   # MCP server configurations
└── skills/                # 43 flat skills (router pattern)
    ├── <skill-name>/      # Each skill is a top-level directory
    │   ├── SKILL.md       # Trigger description + core patterns
    │   └── references/    # Deep content loaded on demand
    └── ...

Agents

Core Development

Operations

Extended

Skills as Commands

All former slash commands have been migrated to skills that auto-trigger based on context:

# Skills auto-trigger — just describe what you want:
"implement user authentication"     # → feature-workflow + authentication
"this endpoint is slow"             # → performance-optimization
"commit these changes"              # → git-workflows
"refactor this function"            # → refactoring
"add docs to this module"           # → documentation
"scan for security issues"          # → owasp
"switch to brainstorm mode"         # → mode-switching

Skills (43 Total)

Every skill follows a flat directory structure with a router pattern: short SKILL.md for triggering + references/ for deep content loaded on demand. Skills auto-trigger based on context — no slash commands needed.

Tech Stack Skills (7 merged routers)

Domain Skills (8)

Pattern Skills (3)

Workflow Skills (7 — migrated from commands)

Methodology Skills (18)

Key methodology principles:

• TDD Strict: No production code without failing test first
• Verification: Evidence-based completion claims
• Quality Gates: Code review between every task
• Bite-sized Tasks: 2-5 minute increments with exact code
• Sequential Thinking: Step-by-step reasoning with confidence scores

Bundled Resources

Skills include progressive-disclosure resources loaded on demand:

Behavioral Modes

Switch modes to optimize responses for different task types:

"switch to brainstorm mode"   # → mode-switching skill activates
"let's focus on implementation" # → implementation mode

Token Optimization

Reduce costs by 30-70% with compressed output modes:

MCP Integrations

MCP servers extend Claude Kit with powerful capabilities. They are automatically used when configured.

How MCP Servers Enhance Skills

Example: Full Feature Development

"Add user profile with avatar upload"

1. feature-workflow skill activates → orchestrates the workflow
2. Context7 → Fetches latest React/Next.js file upload docs
3. Sequential → Plans component structure step-by-step
4. Memory → Recalls your UI patterns from previous sessions
5. Playwright → Tests the upload flow in browser

Setup: See .claude/mcp/README.md

Customization

CLAUDE.md

The .claude/CLAUDE.md file is your project context. Customize it with:

# Project: Your Project Name

## Tech Stack
- **Backend**: FastAPI
- **Frontend**: Next.js
- **Database**: PostgreSQL

## Conventions
- Use type hints
- 80% test coverage
- Conventional commits

## Agent Overrides
### Tester
- Framework: pytest
- Coverage: 90%

Adding Custom Skills

Create a new skill in .claude/skills/<name>/SKILL.md:

---
name: my-skill
description: >
  What this skill does and when to trigger it. Be specific — list
  contexts, keywords, and scenarios. 2-4 pushy sentences.
---

# My Skill

Brief overview.

## When to Use
- Scenario 1
- Scenario 2

## When NOT to Use
- Anti-trigger scenario

---

## Core Patterns
### Pattern Name
Code examples with good/bad comparisons.

## Best Practices
## Common Pitfalls
## Related Skills

Optionally add bundled resources:

my-skill/
├── SKILL.md
├── references/    # Loaded into context on demand
├── scripts/       # Executed without loading into context
└── templates/     # Scaffolded into user's project

Workflow Chains

Skills chain automatically based on context. Here are common flows:

Feature Development

brainstorming → writing-plans → feature-workflow → requesting-code-review → git-workflows

Bug Fix

systematic-debugging → root-cause-tracing → test-driven-development → verification-before-completion

Ship Code

verification-before-completion → requesting-code-review → git-workflows → finishing-a-development-branch

Superpowers Workflow (Detailed)

brainstorming → writing-plans → executing-plans → git-workflows

Uses one-question-at-a-time design, 2-5 min tasks with exact code, subagent execution with code review gates.

Parallel Work

dispatching-parallel-agents → subagent-driven-development → verification-before-completion

Launch multiple agents for independent tasks, then verify results.

Cost-Optimized Session

"switch to token-efficient mode" → [work on tasks] → "switch to default mode"

Enable compressed outputs for high-volume sessions.

Requirements

• Claude Code 1.0+
• Git
• Node.js or Python (depending on your stack)

License

MIT

Built with duthaho