Customizing
Claude Code

What goes in it?

• Tech stack & architecture
• Common commands (test, build, lint)
• Code conventions & style rules
• Project-specific patterns

Where to put it

• ./CLAUDE.md: project (shared)
• ./CLAUDE.local.md: personal
• ~/.claude/CLAUDE.md: global
• .claude/rules/*.md: modular

Where they live

Personal (all projects):
~/.claude/skills/name/SKILL.md

Project (shared via git):
.claude/skills/name/SKILL.md

Key options

• user-invocable: you can call it
• disable-model-invocation: only manual
• argument-hint: autocomplete hint
• allowed-tools: skip permissions
• context: fork: isolated sub-agent

Official

DOCX, PDF, PPTX, XLSX, Web Artifacts, Skill Creator, TDD, Playwright

Creative

Algorithmic Art, Canvas Design, Image Generation, Presentations, GIF Creator

Developer

Superpowers (20+ skills), 65 Full-Stack Skills, Git Worktrees, MCP Builder

Science

125+ Scientific Skills, 77 AI Research Skills, Materials Simulation, Health

Partners

Cloudflare, Google Labs, Hugging Face, Supabase, Notion, HashiCorp

Productivity

Brand Guidelines, Internal Comms, Twitter Optimizer, Reddit Research

Superpowers Collection

20+ battle-tested skills: TDD workflows, /brainstorm, /write-plan, /execute-plan, debugging, and collaboration patterns.

Office Documents

Create and edit Word, Excel, PowerPoint, and PDF files. Fill forms, add charts, generate slides from prompts.

Cloudflare (6 skills)

Build AI agents, manage Workers/KV/R2, audit Web Vitals, and deploy with Wrangler from conversation.

VibeSec Security

OWASP Top 10 scanning, security checklists for 20+ languages. "If you build web apps, do yourself a favor."

Connect (500+ Apps)

Send emails, create GitHub issues, post to Slack, manage Notion. Real actions across 1000+ services.

Skill Seekers

Point it at any documentation website and it automatically converts it into a Claude skill. The skill that creates skills.

Remote (cloud-hosted)

# GitHub
claude mcp add \
  --transport http github \
  https://api.githubcopilot.com/mcp/

# Then authenticate:
/mcp

Local (your machine)

# PostgreSQL
claude mcp add \
  --transport stdio mydb \
  -- npx -y @bytebase/dbhub \
  --dsn "postgresql://..."

🐙

GitHub

PRs, issues, reviews, code search

🚨

Sentry

Error monitoring, stack traces

🗃

PostgreSQL

Query databases naturally

📝

Notion

Knowledge base access

💬

Slack

Messaging & notifications

💳

Stripe

Payment data & management

Creative & 3D

Blender, Figma, Maya, Unity, Excalidraw. Describe a scene and watch it build in real time.

Music & Media

Spotify, YouTube, text-to-speech. Control your soundtrack and content from conversation.

Physical World

3D printers, NVIDIA robotics, AutoCAD. AI that reaches beyond the screen into atoms.

Data & Knowledge

Obsidian, Knowledge Graphs, Notion. Your AI remembers context across sessions.

Museums & Culture

Met Museum (470K+ artworks), Smithsonian, Open Library. Explore humanity's collections.

Science & Niche

Orbital mechanics, healthcare FHIR, Kaggle datasets, handwriting-to-LaTeX, Anki flashcards.

Blender MCP

Describe a 3D scene in plain English and watch Blender build it in real time. Create rooms, characters, animations.

Puppeteer

Full browser automation: navigate websites, fill forms, take screenshots, generate PDFs. A robot intern that browses for you.

3D Printer Monitor

Ask "how's my print?" and get live status, webcam snapshots, and remote control of your 3D printer.

Obsidian Vault

Link your AI to your notes. Query your second brain, search by tag, update notes, and build knowledge conversationally.

Pipedream (2,500+ APIs)

Connect to thousands of APIs with 8,000+ prebuilt tools through a single MCP. The swiss army knife of integrations.

Magg (Meta-MCP)

A meta-MCP that lets AI autonomously discover and install other MCP servers. Your AI extends its own capabilities on demand.

Auto-format after edits

"PostToolUse": [{
  "matcher": "Edit|Write",
  "hooks": [{
    "type": "command",
    "command": "jq -r '.tool_input.file_path'
      | xargs prettier --write"
  }]
}]

Block editing .env files

"PreToolUse": [{
  "matcher": "Edit|Write",
  "hooks": [{
    "type": "command",
    "command": "jq -r '.tool_input.file_path'
      | grep -q '.env'
      && exit 2 || exit 0"
  }]
}]

Desktop notification

"Notification": [{
  "matcher": "permission_prompt",
  "hooks": [{
    "type": "command",
    "command": "osascript -e
      'display notification
      \"Claude needs attention\"'"
  }]
}]

Log all bash commands

"PostToolUse": [{
  "matcher": "Bash",
  "hooks": [{
    "type": "command",
    "command": "jq -r '.tool_input.command'
      >> ~/.claude/cmds.log"
  }]
}]

Plugin structure

my-plugin/
├── .claude-plugin/
│   └── plugin.json    ← manifest
├── skills/
│   └── review/SKILL.md
├── agents/
│   └── analyzer.md
├── hooks/
│   └── hooks.json
└── .mcp.json          ← integrations

Minimal manifest

{
  "name": "my-plugin",
  "description": "What it does",
  "version": "1.0.0"
}

Test locally:

claude --plugin-dir ./my-plugin

# Skills are namespaced:
/my-plugin:review

Permission rules

{
  "permissions": {
    "allow": [
      "Bash(npm run *)",
      "Bash(git *)",
      "Read(./src/**)"
    ],
    "deny": [
      "Bash(rm *)",
      "Read(.env*)"
    ]
  }
}

Settings file priority

Interactive setup: type /config

Start simple

• Run /init to create a CLAUDE.md
• Add one MCP server you use daily
• Create one skill for a repetitive task

Scale up later

• Add hooks for formatting & protection
• Use .claude/rules/ for organized guidelines
• Bundle into a plugin for your team