Analysis updated 2026-05-18
Automatically hand backlog issues to Codex for implementation while Claude verifies and merges.
Run an unattended, self pacing development loop that restarts itself without manual kicks.
Track the state of every AI assigned task using GitHub issue labels instead of in-memory context.
| bostonvex/claude-codex-orchestrator | abhas9/escape-run | abhishek-kumar09/mern-live-js-june-2021 | |
|---|---|---|---|
| Stars | 1 | 1 | 1 |
| Language | HTML | HTML | HTML |
| Last pushed | — | — | 2021-09-13 |
| Maintenance | — | — | Dormant |
| Setup difficulty | moderate | easy | moderate |
| Complexity | 4/5 | 2/5 | 2/5 |
| Audience | developer | general | developer |
Figures from each repo's GitHub metadata at analysis time.
Requires an authenticated gh CLI with merge and push permissions, plus a Codex CLI subscription or OpenAI API key for local workers.
codex-loop is a tool that turns Claude Code into a self running work loop for any code repository. It splits a backlog of tasks between two AI coding assistants: Claude, which plans, checks, merges, and ships the work, and Codex, which implements the tasks it is assigned. Every task is tracked as a normal GitHub issue, and the labels plus a small set of comment conventions on that issue act as the memory of where each task stands, so nothing about the state depends on the AI remembering anything between runs. The main problem it solves is that a typical automated loop stops as soon as its work queue is briefly empty or one pass finishes, and a person has to manually start it again. codex-loop instead paces itself: it checks more frequently while work is actively moving, checks less often when there is nothing to do, and stops entirely once everything is finished or paused, without needing anyone to restart it by hand. Using it requires Claude Code itself, the gh command line tool logged into the target repository, and git. By default it also needs the codex command line tool, which runs on Node.js and is billed through its own ChatGPT or Codex subscription or an OpenAI API key rather than through Claude or Anthropic. Depending on setup, work can also be sent to a Codex Cloud agent instead of running Codex locally. The account running all of this needs permission in the target repository to create labels and issues, merge pull requests, and push to the default branch. First use in a repository triggers an automatic setup step that creates the needed labels and a single pinned tracking issue holding its configuration, and every setting in that configuration has a working default so nothing has to be changed before a first run. From there, a person opens GitHub issues for each piece of work, labels each one for either Claude or Codex to implement, and either runs one iteration by hand or starts the self pacing autonomous mode. The loop never force pushes and never merges when checks are failing.
A Claude Code skill that runs an autonomous work loop, splitting tasks between Claude and Codex and tracking every task as a self pacing GitHub issue.
Mainly HTML. The stack also includes Claude Code, Codex CLI, GitHub CLI.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.