keepitmello/claude-codex-meight

6 stars · Last commit 2026-08-01

Agent-first harness: Claude Code drives OpenAI Codex workers like native subagents — mid-turn steering, token-free observation, bidirectional QUESTION protocol. Built on the official openai-codex Python SDK.

README preview

# claude-codex-meight

<p align="center">
  <img src="./docs/hero.jpg" alt="Claude Fable 5 + Codex" width="720">
</p>

**English** | [한국어](./docs/README.ko.md)

> **A two-way harness where Codex mates challenge your plans and Codex workers
> build them.** Meight is built for an LLM agent that designs collaboratively, delegates,
> steers, reviews, and signs off with evidence. Official `openai-codex` Python
> SDK underneath. CLI: `meight`.

Most bridges are built for a human watching a terminal: tmux panes, dashboards,
stdout scraping. Meight is built for the agent itself — a dispatcher that
starts hidden Codex sessions, reads compact disk digests, steers a live turn,
answers structured questions, and keeps final reports small enough to make
user-facing decisions without drowning in implementation detail.

The core idea is that a frontier model wasted as a silent executor is capability

View full repository on GitHub →