Claude Code Agent Farm CLI
An orchestrator that runs 20 or more Claude Code sessions in tmux in parallel to fix bugs or roll out practices across a codebase
- 923 ★ on the repository
- Last commit
Broad access
Why this level
- Dozens of agents run with `--dangerously-skip-permissions`, edit files and make commits without permission prompts.
- Parallel sessions burn through Claude Code subscription limits quickly.
Install
- Works in
Manual install
git clone https://github.com/Dicklesworthstone/claude_code_agent_farm.git && cd claude_code_agent_farm && chmod +x setup.sh && ./setup.shThe quick-start commands from the README; the script sets up the environment and the `cc` alias.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Dozens of agents run with `--dangerously-skip-permissions`, edit files and make commits without permission prompts.
- Parallel sessions burn through Claude Code subscription limits quickly.
This is third-party code. Review the repository files before installing.
What it does
Claude Code Agent Farm launches many `claude` sessions in tmux panes and hands them work on a single project. The configuration sets the agent count: 20 or more by default, with a ceiling of 50 through the `max_agents` parameter. There are three modes: bug fixing, rolling out best practices from a guide for the chosen stack, and a cooperating-agents mode with file locks against conflicts. Configurations are ready for 34 stacks, from Next.js and Python to Rust, Go and Flutter, with 37 prompts. The orchestrator watches the agents: it shows a status dashboard, warns about full context, restarts stuck ones, makes commits and writes an HTML report. It starts with `claude-code-agent-farm --path <project> --config <config>`.
Who it is for. Developers with a large codebase and a Claude Code subscription who are ready to run agents on a wide front and review the commits afterwards.
Good fit when
- A Next.js project has accumulated hundreds of type and lint errors that suit parallel cleanup.
- A repository has to be brought in line with a stack best-practices guide by splitting the work into chunks.
- Several agents work on one feature without overwriting each other's files, through the lock system.
- The environment has to be verified before launch with `claude-code-agent-farm doctor --path <project>`.
Not a fit when
- The repository has no backup or branch to roll back mass commits.
- A single careful session with manual approval of every action is needed.
- Subscription limits or the token budget are not sized for dozens of simultaneous sessions.
Example request
Run claude-code-agent-farm for this Next.js project with configs/nextjs_config.json on 10 agents and show the monitoring dashboard.Limitations
Launch goes through the `cc` alias with the `--dangerously-skip-permissions` flag, so agents work without permission prompts. Python 3.13, tmux, git and an installed Claude Code are required. Dozens of sessions burn through subscription limits quickly. Monitoring is tailored to Claude Code: the welcome banner and context reset work with caveats for other CLIs such as OpenCode or Codex. The license is MIT with an OpenAI and Anthropic rider. The last commit is from September 2026.
How to disable. Stop the orchestrator with a double Ctrl+C within three seconds for a forced shutdown and close the tmux session. Then delete the project folder and the `cc` alias from the shell configuration.
FAQ
How do I install Agent Farm?
Clone the repository and run `./setup.sh`. The script checks prerequisites, creates a Python 3.13 environment and configures the `cc` alias.
Can other agents run instead of Claude Code?
Yes, the orchestrator only types `cc` into each tmux pane, so the alias can point at OpenCode or Codex CLI. Monitoring is built for Claude Code and works with limits.
How are conflicts between agents avoided?
The cooperating-agents mode uses a file lock system. In the regular modes work is split into chunks and progress is recorded in commits with a short change summary.
What does the monitoring dashboard show?
The state of each agent, warnings about full context and activity heartbeats. Ctrl+R broadcasts `/clear` to every agent at once.
What is left after a run?
Git commits with change summaries and an HTML run report. Launch parameters live in a JSON configuration that can be adapted to your stack.
Discussion
No questions or reviews yet.
Related
CLI toolsA terminal app for running several Claude Code, Codex, Gemini CLI and Aider sessions in parallel, each in its own tmux session and git worktree
Claudette runs several Claude Code sessions in parallel, each in its own git worktree, from a window or the console
Multi-agent orchestration for Claude Code: teams, roles, autopilot and task routing across skills and subagents
Workflows for Claude Code: /do, /omo, /bmad-pilot and more commands, with Codex, Gemini or OpenCode writing the code through codeagent-wrapper
A tmux-based TUI that runs Claude Code, Codex and other agents in one list: status, quick prompts, worktrees and diff review
A template for Claude Code projects: parallel agent swarms, workflow skills, safety hooks and rules in one install.
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| Claude Code Agent Farm (this card) | no | 923 | |
| Claw Orchestrator | no | 606 | orchestrates several agents as a CLI; here the focus is 20+ parallel Claude Code sessions and ready stacks |
| tmux-ide | yes | 567 | turns tmux into a development environment for agents; here an orchestrator for mass edits driven by a stack configuration |
README in short
The README lists requirements, a four-step quick start, 34 supported stacks and configuration parameters. Separate sections cover the monitoring dashboard, automatic agent restarts and HTML run reports.