#194194th of 218 in Coding agents and AI editors

claude_code_agent_farm

Runs parallel Claude Code sessions in tmux to fix lint errors and apply stack guides

Stars
923
License
custom license
Last commit
Sep 2026
Language
Shell
Stars, last 29 weeks+148
22 Apr 2026: 775 stars29 Apr 2026: 789 stars6 May 2026: 795 stars13 May 2026: 804 stars20 May 2026: 827 stars27 May 2026: 829 stars3 Jun 2026: 831 stars10 Jun 2026: 833 stars17 Jun 2026: 835 stars24 Jun 2026: 838 stars1 Jul 2026: 842 stars8 Jul 2026: 845 stars15 Jul 2026: 851 stars22 Jul 2026: 870 stars29 Jul 2026: 873 stars5 Aug 2026: 875 stars12 Aug 2026: 896 stars19 Aug 2026: 906 stars26 Aug 2026: 909 stars2 Sept 2026: 912 stars9 Sept 2026: 912 stars16 Sept 2026: 913 stars23 Sept 2026: 915 stars30 Sept 2026: 916 stars7 Oct 2026: 917 stars14 Oct 2026: 918 stars21 Oct 2026: 918 stars28 Oct 2026: 920 stars4 Nov 2026: 921 stars11 Nov 2026: 923 stars

775 on 22 Apr 2026923 on 11 Nov 2026

From GitHub's daily star counts.

Checked on GitHub.

Overview

A Python orchestrator that launches 20+ Claude Code sessions (up to 50 via max_agents) in tmux panes. It feeds them chunks of type-checker and linter output, or a per-stack guide to implement step by step, and agents work through them in parallel. A third workflow has agents claim files through lock files in a /coordination/ directory. It ships 34 stack configs, per-stack prompts, a monitoring dashboard with auto-restart, and tool setup scripts.

Who it is for: Developers running many parallel Claude Code agents on a large codebase

Strengths

  • Ships 34 per-stack configs with prompts and 31 stack guides
  • Monitors context use and idle state, restarts agents, writes JSON monitor state
  • Cooperating workflow uses lock files and a work registry, with stale-lock handling after 2 hours
  • doctor command checks the setup before a run; config values are overridable from the CLI

Weaknesses

  • Requires launching Claude Code with --dangerously-skip-permissions via the cc alias
  • Health checks and context resets assume Claude Code; other CLIs get degraded support
  • Conflict avoidance in the cooperating workflow comes from the prompt, not enforced code
  • Custom license (MIT plus OpenAI/Anthropic rider), detected as NOASSERTION

What it needs

  • no GPU
  • Needs Python 3.13+, uv, tmux, git, Claude Code CLI, direnv (optional)
  • Models: Claude Code, OpenCode (limited), OpenAI Codex CLI (limited)

Also in Coding agents and AI editors

See all 218
Also in Coding agents and AI editors
RankProjectScore
1claw-codeRust CLI coding agent harness using Anthropic and OpenAI-compatible APIs194.8k stars+724 (+0.4%) in 30 daysMIT96 out of 100
2t3codeDesktop, web and mobile front end for controlling local coding agent CLIs Live demo ↗ (opens in a new tab)57.9k stars+30,427 (+111%) in 30 daysMIT95 out of 100
3codewhaleTerminal coding agent in Rust that works with 40+ model providers41.2k stars+331 (+0.8%) in 30 daysMIT94 out of 100
#4ArchonYAML workflow engine that runs AI coding agents through fixed dev steps23.9k starsMIT93 out of 100
#5piMinimal terminal coding agent extensible with skills and packages121.6k stars+7,287 (+6.4%) in 30 daysMIT92 out of 100