herry2059/project-os-for-codex — explained in plain English
Analysis updated 2026-05-18
Give Codex a persistent, Git backed record of project goals, acceptance criteria, and progress instead of relying on chat history.
Let a new AI session or teammate resume a project using a handoff package instead of rereading old prompts.
Track which tasks are done, blocked, or at risk on a project dashboard.
Grant Codex a scoped, short lived credential instead of full account access.
| herry2059/project-os-for-codex | jazzband/web3-devkit-cli | metanode-trade/hyperliquid-trading-bot | |
|---|---|---|---|
| Stars | 101 | 101 | 101 |
| Language | TypeScript | TypeScript | TypeScript |
| Setup difficulty | moderate | moderate | moderate |
| Complexity | 4/5 | 3/5 | 3/5 |
| Audience | developer | developer | developer |
Figures from each repo's GitHub metadata at analysis time.
Requires running a local server, generating scoped credentials, and connecting Codex through MCP.
Project OS for Codex is an open-source control plane built around OpenAI's Codex agent, meant for the moment a team asks what the AI is doing, where the next work session should pick up, and how another person or AI can take over without rereading every chat. It is not a chat interface itself. Instead it works alongside Codex to track a project's goals, acceptance criteria, progress, risks, and next steps, storing all of it in Git so there is a real record of the work rather than scattered notes in prompts and screenshots. The tool exists because the author noticed a recurring pattern in real client and internal projects: an AI writes code quickly, but progress stays hidden inside chat history, the next session does not know where to resume, humans cannot tell which tasks are actually done or blocked, and handoffs between people or between AI sessions become painful. Project OS turns those scattered pieces, kickoff notes, Git commits, progress updates, and knowledge, into a single structured workspace. To use it with Codex, a project owner creates a short lived credential scoped to a single workspace and project rather than sharing an account password, then runs a generated command that connects Codex through the MCP protocol, a standard way for AI tools to call external services. The connection currently exposes two tools: one that lets Codex read a project's kickoff card, rules, current progress, and next step, and another that lets it append a single verified progress update tied to a matching Git commit. Credentials expire automatically, can be revoked, and cannot touch member management, API keys, deletions, or deployments, keeping the AI's access deliberately narrow. The basic workflow is to create a project with a kickoff card describing the goal, acceptance criteria, what is explicitly out of scope, and the owner, after which the system keeps a Git backed record of progress and issues as work happens. A dashboard shows overall status and health, and a handoff package lets the next person or AI resume the project without reading through the entire history. The project is written in TypeScript and released under the Apache License 2.0.
An open-source control plane that gives OpenAI Codex a Git-backed record of project progress, acceptance criteria, and handoffs between sessions.
Mainly TypeScript. The stack also includes TypeScript, MCP, Codex.
Free to use, modify, and distribute, including commercially, as long as you keep attribution notices, under the Apache License 2.0.
Setup difficulty is rated moderate, with roughly 1h+ to a first successful run.
Mainly developer.
This repo across BitVibe Labs
double-check against the repo, no cap.