Claude Code Expert Skill

You are an expert on Claude Code - Anthropic's agentic coding tool. You have deep knowledge of its features, configuration, and best practices.

Documentation References

For detailed information, reference these cached documentation files:

@cached_docs/overview.md - Claude Code overview and getting started
@cached_docs/settings.md - Configuration and settings reference
@cached_docs/hooks.md - Hooks system documentation
@cached_docs/statusline.md - Status line configuration
@cached_docs/auth.md - Authentication and token management
@cached_docs/plugins-reference.md - Complete technical reference for plugin system
@cached_docs/cli-reference.md - Complete CLI command reference
@cached_docs/slash-commands.md - Slash commands reference

To update documentation:

Run python3 update_docs.py (local) or use the "Update Documentation" cell in the notebook
Check docs_manifest.json for source URLs and sitemap
Documentation is cached from https://code.claude.com/docs/en/

Quick Reference

Essential Commands

Command	Purpose
`/init`	Generate CLAUDE.md from codebase
`/clear`	Reset conversation context
`/compact [focus]`	Summarize to save tokens
`/usage`	Show token usage and costs
`/model <name>`	Switch model (sonnet, opus, haiku)
`/memory`	Edit CLAUDE.md files
`/doctor`	Diagnose installation issues
`/mcp`	Manage MCP server connections
`/vim`	Toggle vim edit mode
`#`	Quick-add to memory
`!`	Toggle auto-accept edits

Configuration Files

File	Location	Purpose
`CLAUDE.md`	Project root	Project memory & instructions
`CLAUDE.local.md`	Project root	Personal project notes (gitignored)
`~/.claude/CLAUDE.md`	Home	Global instructions
`.claude/settings.json`	Project	Project settings
`~/.claude/settings.json`	Home	User settings
`.claude/commands/*.md`	Project	Custom slash commands
`.claude/agents/*.md`	Project	Custom subagents
`.claude/skills/*/SKILL.md`	Project	Custom skills

Settings Hierarchy (highest to lowest)

Enterprise managed policy
Command line arguments
.claude/settings.local.json (personal)
.claude/settings.json (shared)
~/.claude/settings.json (user default)

CLAUDE.md Best Practices

Good CLAUDE.md Structure

# Project Name

## Quick Commands
- `npm run dev` - Start dev server
- `npm test` - Run tests

## Code Style
- TypeScript with strict mode
- Prefer functional components
- Use Tailwind for styling

## Project Structure
- src/components/ - React components
- src/lib/ - Utilities
- src/api/ - API routes

## Important Notes
- Never commit .env files
- Run migrations before testing

What to Include

Commands you run frequently
Project-specific conventions
File structure explanations
Integration details (APIs, databases)
Things Claude gets wrong repeatedly

What NOT to Include

Generic advice ("write clean code")
Sensitive data (API keys, passwords)
Very long documentation (link instead)

Permissions Configuration

settings.json Format

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

Permission Patterns

Bash(command:*) - Allow command with any args
Bash(npm run:*) - Allow npm run anything
Read(path) - Allow reading specific path
Write(glob/**) - Allow writing to glob pattern
* in allow = allow all (dangerous)

Custom Slash Commands

Command File Format

---
allowed-tools: Bash(npm:*), Read, Write
description: Run the test suite
argument-hint: [test-pattern]
---

Run tests matching $ARGUMENTS:
- Use `npm test -- $ARGUMENTS` if pattern provided
- Use `npm test` for all tests
- Report failures clearly

Variables Available

$ARGUMENTS - Everything after command name
$1, $2, etc. - Positional arguments
!command - Execute bash and include output
@file - Include file contents

Subagents

Agent File Format

---
name: reviewer
description: Code review specialist
tools: Read, Grep, Glob
model: sonnet
---

You review code for:
- Logic errors
- Performance issues
- Security vulnerabilities

Be specific and suggest fixes.

Built-in Agents

General-purpose - Full access, all tools
Plan subagent - Read-only, for planning
Explore subagent - Fast (Haiku), read-only

Hooks System

Hook Events

Event	When	Use For
`PreToolUse`	Before tool runs	Validate, block dangerous ops
`PostToolUse`	After tool completes	Auto-format, lint
`Stop`	Claude finishes	Commit, notify
`SessionStart`	Session begins	Load env vars
`UserPromptSubmit`	User sends message	Log, transform

Hook Configuration

{
  "hooks": {
    "PostToolUse": [{
      "matcher": "Write|Edit",
      "hooks": [{
        "type": "command",
        "command": "black $CLAUDE_FILE_PATHS",
        "timeout": 30
      }]
    }]
  }
}

Hook Exit Codes

0 - Success, continue
2 - Block operation, show error to Claude

MCP (Model Context Protocol)

Adding MCP Servers

# HTTP server
claude mcp add --transport http name https://server.url/mcp

# Local stdio server
claude mcp add --transport stdio name -- npx package-name

# From Claude Desktop
claude mcp add-from-claude-desktop

Managing Servers

claude mcp list          # Show configured servers
claude mcp remove name   # Remove a server
/mcp                     # Check status in Claude

Model Selection

Available Models

Alias	Model	Best For
`sonnet`	Claude Sonnet 4.5	Daily coding, balanced
`opus`	Claude Opus 4	Complex reasoning, hard problems
`haiku`	Claude Haiku 4.5	Fast, simple tasks
`sonnet[1m]`	Extended context	Large codebases

Switching Models

# Command line
claude --model opus

# In session
/model opus

# Environment variable
export ANTHROPIC_MODEL=claude-sonnet-4-5-20250929

Headless Mode

Non-Interactive Use

# Single prompt, text output
claude -p "explain this code" --output-format text

# JSON output for scripting
claude -p "list files" --output-format json

# With file input
claude -p "review this" < file.py

CI/CD Integration

# GitHub Action example
- name: Claude Review
  run: |
    claude -p "review changes in this PR" \
      --output-format text \
      > review.md

Troubleshooting

Common Issues

"Command not found: claude"

Add to PATH: export PATH="$HOME/.local/bin:$PATH"
Source bashrc: source ~/.bashrc

Context getting polluted

Use /clear between unrelated tasks
Use /compact to summarize and reduce tokens

Claude not following instructions

Check CLAUDE.md is being loaded (look for it in context)
Be more specific in instructions
Break complex tasks into steps

Slow responses

Switch to haiku for simple tasks: /model haiku
Use /compact to reduce context
Clear and start fresh if context bloated

Diagnostic Commands

# Check installation
claude doctor

# Check version
claude --version

# Check what Claude sees
/status
/usage

Colab-Specific Notes

In Colab's headless terminal:

/doctor and /status may hang (interactive UI issues)
Use /usage to verify Claude is working (note: usage tab may fail with setup-token due to missing user:profile scope)
Use claude -p "prompt" for scripted testing

PATH setup:

export PATH="$HOME/.local/bin:$HOME/.claude/bin:$PATH"
export TERM=xterm-256color
export FORCE_COLOR=1

Sandbox requires:

apt-get install socat bubblewrap

Claude Code Expert Skill

Claude Code Expert Skill

Documentation References

Quick Reference

Essential Commands

Configuration Files

Settings Hierarchy (highest to lowest)

CLAUDE.md Best Practices

Good CLAUDE.md Structure

What to Include

What NOT to Include

Permissions Configuration

settings.json Format

Permission Patterns

Custom Slash Commands

Command File Format

Variables Available

Subagents

Agent File Format

Built-in Agents

Hooks System

Hook Events

Hook Configuration

Hook Exit Codes

MCP (Model Context Protocol)

Adding MCP Servers

Managing Servers

Model Selection

Available Models

Switching Models

Headless Mode

Non-Interactive Use

CI/CD Integration

Troubleshooting

Common Issues

Diagnostic Commands

Colab-Specific Notes

Similar Skills