gptme-cc-plugin

Claude Code skills for gptme — use gptme's terminal AI agent directly from your Claude Code sessions.

What is this?

This package adds three Claude Code skills that let you delegate work to gptme without leaving your CC session:

Skill	Invocation	What it does
gptme-run	/gptme:run "<task>"	Delegate a task to gptme's agent (shell, files, browser)
gptme-review	/gptme:review	Multi-lens code review (correctness, security, test coverage)
gptme-context	/gptme:context	Inject gptme workspace lessons/memory into this CC session

Prerequisites

• gptme installed and in PATH:

  pip install gptme
  # or
  pipx install gptme
  

• Claude Code with Bash tool permissions enabled

Install

git clone https://github.com/gptme/gptme-cc-plugin ~/.claude/skills/gptme

That's it. Restart Claude Code and invoke /gptme:run, /gptme:review, or /gptme:context.

Usage

Delegate a task

/gptme:run "find all TODO comments in this repo and create GitHub issues for each"

Multi-lens code review

/gptme:review

Runs gptme with three independent review passes (correctness, security, test coverage) on your current git diff HEAD.

Load workspace context

/gptme:context

Injects lessons, skills, and memory from the current gptme workspace into this Claude Code session.

Why?

The pattern was proven by OpenAI's codex-plugin-cc: packaging a competing agent as a CC skill reaches CC users with zero friction. A gptme plugin is a distribution play — CC users who try gptme via a skill become gptme users.

Architecture

These are skills-based (SKILL.md) rather than MCP-based. Each skill teaches Claude Code to invoke the local gptme binary via the Bash tool. No server required.

Future work: when gptme ships an MCP server mode, a Phase 2 MCP-based version will expose gptme.run and gptme.review as structured tools.

Related

• gptme — the agent these skills delegate to
• gptme-agent-template — fork to create your own gptme workspace
• Bob — an example of a rich gptme workspace (the agent that built this)

License

MIT