Claude Code (Official CLI)

This skill wraps the @anthropic-ai/claude-code tool. It provides a powerful, agentic coding experience directly in your terminal.

🛠️ Prerequisites

1. Install:

   npm install -g @anthropic-ai/claude-code
   # Or macOS (Homebrew):
   # brew install --cask claude-code
   

2. Authenticate (Run once manually):

   claude login
   

   Follow the browser flow to connect your Anthropic account.

🚀 Usage Patterns

Always use pty:true for claude commands to ensure proper interaction and output formatting.

1. One-Shot Task (The -p flag)

Use the -p (print/prompt) flag for single, non-interactive tasks or when piping input. This is best for automation.

# Direct task
bash pty:true workdir:~/my-project command:"claude -p 'Fix the bug in auth.ts where the token expires too soon'"

# Code Review (Pipe git diff)
bash pty:true workdir:~/my-project command:"git diff main | claude -p 'Review these changes for security vulnerabilities'"

2. Interactive Session

Start a persistent session where you can give follow-up instructions.

# Start session
bash pty:true workdir:~/my-project command:"claude"

# In the session (via process tool):
# > Write input: "Refactor the database schema"
# > Write input: "/compact" (to save context)

3. Project Configuration (CLAUDE.md)

Create a CLAUDE.md file in your project root to teach Claude about your project's style, build commands, and architecture.

# CLAUDE.md
- Build command: `npm run build`
- Test command: `npm test`
- Style: Use TypeScript, functional components, and Tailwind CSS.

🎛️ Key Commands & Flags

Command / Flag	Description
claude	Start interactive session
claude -p "text"	Run one-shot prompt (Recommended for agents)
/compact	Reduce context window usage (in session)
/cost	Show current session cost (in session)
/init	Initialize a new project with CLAUDE.md
/bug	Report a bug to Anthropic

⚠️ Best Practices for Agents

1. Use -p for reliability: When automating, claude -p "prompt" is more predictable than interacting with the shell REPL.
2. Safety: Claude Code can execute shell commands. Ensure you are running in a safe environment or monitoring its output.
3. Context: Always run from the root of the git repository (workdir:~/project) so Claude sees the full context.
4. Auto-Updates: The npm package updates frequently. Run npm update -g @anthropic-ai/claude-code regularly.

🤖 Example Workflow: Feature Implementation

1. Plan:

   bash pty:true command:"claude -p 'Read src/api and plan a rate-limiting feature'"
   

2. Implement:

   bash pty:true command:"claude -p 'Implement the rate-limiting plan using Redis'"
   

3. Verify:

   bash pty:true command:"claude -p 'Write a test for the rate limiter and run it'"