code-any

中文版

A terminal-based coding agent built in TypeScript — a Claude Code-like workflow you can run locally.

What is this?

This project demonstrates how a Claude Code-style agent works under the hood:

• Agent loop — iterative tool-use rounds with the Anthropic API
• Context engineering — budget-aware truncation so large outputs don't flood the model
• Explore subagent — a cheap read-only pass to gather context before the main agent acts
• Tool output separation — raw output stays local, only structured summaries go to the model

If you've ever wondered "how does Claude Code actually work?", read the source.

Quick Start

# Install
npm install && npm run build && npm link

# Configure
cp .env.example .env
# Set ANTHROPIC_AUTH_TOKEN=your_token_here in .env

# Run
code-any

CLI Flags

code-any --model claude-3-7-sonnet-latest  # choose model
code-any --cwd /path/to/workspace          # set working directory
code-any --yolo                            # auto-approve all tool calls

Slash Commands

Command	Description
/help	Show built-in commands
/tools	List available tools
/model	Show current model
/approval	Show approval mode
/diff	Show workspace diff
/clear	Clear conversation
/exit	Exit

Built-in Tools

Tool	Description
list_files	List files in a directory
read_file	Read file content (with slicing)
search_code	Ripgrep-based code search
write_file	Write/create files (needs approval)
run_shell	Execute shell commands (needs approval)
diff_workspace	Show workspace changes

Configuration

ANTHROPIC_AUTH_TOKEN=your_token_here        # required
ANTHROPIC_MODEL=claude-3-7-sonnet-latest    # main model (default)
EXPLORE_MODEL=claude-3-5-haiku-latest       # explore subagent model (default)
DEFAULT_APPROVAL=default                    # or "never" for yolo mode

For Developers

npm run dev        # Run in development mode
npm test           # Run tests
npm run typecheck  # Type check only
npm run build      # Build bundle

License

MIT