gptme-claude-code

Claude Code plugin for gptme - spawn Claude Code subagents for various coding tasks.

Features

This plugin allows gptme agents to leverage Claude Code (the claude CLI) as subagents for:

Analyze: Code reviews, security audits, test coverage analysis
Ask: Answer questions about codebases
Fix: Fix lint errors, build issues, type errors
Implement: Implement features in isolated worktrees

Installation

Requires Claude Code CLI (claude) to be installed:

npm install -g @anthropic-ai/claude-code

Then install this plugin:

pip install gptme-claude-code

Usage

The plugin provides several functions available via ipython:

analyze() - Code Analysis

# Quick security scan
analyze("Review this codebase for security vulnerabilities.")

# Code review
analyze("Review changes in the last commit for code quality issues.")

# Test coverage analysis
analyze("Analyze test coverage and identify critical untested paths.")

ask() - Code Questions

# Understand code structure
ask("How does the authentication flow work in this codebase?")

# Find implementations
ask("Where is the database connection pool configured?")

fix() - Fix Issues

# Fix lint errors
fix("Fix all mypy type errors in src/")

# Fix build issues
fix("The tests are failing with ImportError, diagnose and fix.")

implement() - Implement Features

# Simple implementation
implement("Add a --verbose flag to the CLI")

# Complex implementation (uses worktree for isolation)
implement("Implement rate limiting for the API endpoints", use_worktree=True)

Background Tasks

For long-running tasks, use background=True:

result = analyze("Comprehensive security audit", background=True, timeout=1800)
# Returns session ID

check_session("claude_code_a1b2c3d4")  # Check progress
kill_session("claude_code_a1b2c3d4")   # Cancel if needed

Configuration

The plugin requires:

Claude Code CLI installed (npm install -g @anthropic-ai/claude-code)
Anthropic API key configured (usually via Claude Code's own auth)

Cost Efficiency

Claude Code uses a subscription model ($200/mo for Pro), making it cost-effective for parallel analysis tasks compared to direct API calls.

When to Use

Use Claude Code subagents for:

Single-purpose analysis (no gptme tool ecosystem needed)
Parallel analysis tasks
Tasks that benefit from fresh context (no accumulated history)
Long-running analysis where background mode is appropriate

Use gptme tools directly for:

Complex multi-step workflows
Tasks requiring file modifications with review
Interactive debugging sessions
Tasks needing gptme's full context

License

MIT