chord CLI
Run a cluster of agents in one repository: each gets its own worktree, a file claims board and messaging between agents.
- 63 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- Agents edit code in worktrees and `finish` merges branches into main; set up the `test` check.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/chordDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Agents edit code in worktrees and `finish` merges branches into main; set up the `test` check.
This is third-party code. Review the repository files before installing.
What it does
chord starts as many console agents as you like and gives each its own git worktree. A shared board records who holds which file. An agent that needs a file another agent holds asks the holder first and waits, and a hook rejects writes to unclaimed files. Idle agents wake when a message arrives. The `finish` command rebases the branch onto main, fast-forwards it, releases the claims and tells the others in one step. On a conflict nothing changes and the agent is told to rebase and retry. Claude Code, Codex CLI, OpenCode and pi are supported. `chord ls`, `messages` and `events` give an overview, while `stop` and `clean` wind things down.
Who it is for. Developers who run several agents in one repository and are tired of merge conflicts and overwritten files.
Good fit when
- Two agents are launched on neighboring features and touch one shared file.
- Three agents keep separate branches and merging into main should not need manual work.
- A test run such as `npm test` fires on each branch before merging.
- An agent was closed and later you return to the same worktree and conversation.
- A log is needed: who claimed what and where a conflict happened.
Not a fit when
- One agent works on one repository and a plain terminal is enough.
- Node older than 24 is all that is available.
- The repository does not fit one project folder or paths exceed 190 characters.
Example request
Start two agents in the project: one edits the API, the other the docs. Let them agree on a shared file and merge into main when ready.Limitations
Node 24 and git are required, and on Linux also python3, make and g++ to build `node-pty`, with installation using `--allow-scripts=node-pty`. It runs on Windows 11, macOS and Linux, one repository per project. The project expects the repository in a child folder next to the `chord/` folder. Gemini CLI, Cursor CLI, Copilot CLI, Qwen Code and DeepSeek Harness are only planned. `finish` merges a branch into main and `chord clean` removes worktrees and branches of departed agents, keeping unmerged work unless `--force` is given.
How to disable. Stop the agents with `chord stop <name>`, then remove worktrees and branches with `chord clean`. Uninstall the global `chord-cli` package and the project's `chord/` folder.
FAQ
How do agents learn the chord rules?
The rules come from chord itself at start-up and on every prompt. You do not explain them in the task: just hand out work in plain language.
What goes into config.json?
The `setup` and `test` fields. `setup` runs once in every new worktree, for example `npm ci`, and `test` runs after rebasing before main moves.
What does the benchmark show?
On CooperBench with the weaker model chord holds 63% against 34% for the lead-agent mode. With the stronger model it stays near the best setups.
How to return to a departed agent?
Run `chord resume <name>`: it reopens the same worktree and the agent's own conversation for Claude, Codex, OpenCode and pi.
Where does chord write settings?
To the project's `chord/config.json`. chord does not touch `~/.claude.json`, and Codex asks once to review the hooks.
Discussion
No questions or reviews yet.
Related
CLI toolsA tmux-based TUI that runs Claude Code, Codex and other agents in one list: status, quick prompts, worktrees and diff review
multi-agents runs a team of coding agents on a task and measures whether it beat a single agent and what it cost.
DevChain runs a team of AI coding agents on your own machine with a shared board, chat, code review and live terminals.
A bridge between local CLI agents: Codex, Cursor or Claude Code delegate work to Antigravity, Grok, Kimi and seven more.
A CLI for AI agents sharing a repository: a task board, a spec and acceptance by a second agent against a specific commit
One config for Claude Code and Codex in every repo: shared rules, skill packs, second-model review and a destructive-command guard
README in short
The README explains the claims board, inter-agent messages and `finish`, and gives a CooperBench benchmark and the supported agents.