hiroyuki0504

Agents Kit — AI skill for Claude Code

AI community

Conflict-free parallel development on one Git repo by multiple AI coding sessions (Claude Code / Codex).

How to install Agents Kit

This entry records only its repository, not the path inside it, so there is no exact command to give. Open hiroyuki0504/agents-kit and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Agents Kit does

Conflict-free parallel development on one Git repo by multiple AI coding sessions (Claude Code / Codex). Git-only coordination: CAS via push fast-forward on a state branch, last-instruction-wins directives, serialized merges with full tests per merge.

Alternatives in AI

  • Ccstatusline — by sirmalloc - A highly customizable status line formatter for Claude Code CLI that displays model info, git b 5.5k ★
  • Part 4 — Generate AGENTS.md And AI Agent Configuration Files — I'll help you create the instruction files that will guide your AI coding assistant to build your MVP 2.1k ★
  • Oil UI — Push AI UI design to its limits: explore distinct directions, compare them side by side, and refine against re 861 ★

README

agents-kit — 複数AIセッション並行開発キット

[![tests](https://github.com/hiroyuki0504/agents-kit/actions/workflows/tests.yml/badge.svg)](https://github.com/hiroyuki0504/agents-kit/actions/workflows/tests.yml)

**English** — agents-kit lets multiple AI coding sessions (Claude Code / Codex, any mix of models) implement in parallel on **one Git repository without conflicts**. Coordination runs on git alone, with no AI-vendor-specific features: an orphan `agent-state` branch on origin is the shared ledger (git push's fast-forward check acts as compare-and-swap), user instructions are recorded as totally-ordered directives where **the newest instruction wins**, and every merge to main goes through a single serialized `agents merge` path — fresh worktree, merge simulation, symbol-loss check, and a full test run per merge. A local pre-push hook physically blocks direct pushes to main.

Quick start: `./install.sh /path/to/your-repo`, set `test_cmd` in `.agents/config.json`, then tell each AI session **"Read .agents/PROTOCOL.md and follow it."** — that's all.

Documentation and CLI messages are in Japanese. The primary readers are AI agents, which handle this natively; this README covers everything a human operator needs.


1つの Git リポジトリに対し、別々のターミナルで動く複数の AI(Claude Code / Codex 混在可)が、あなたの随時指示を受けながら**コンフリクトを起こさず並行実装**するための調整キットです。

仕組みは git だけで完結します(AI 固有機能に依存しない)。調整情報は origin 上の専用ブランチ `agent-state` に記録され、main へのマージは直列化された `agents merge` コマンドだけが行います。main への直接 push はローカルフックが物理的に拒否します。

動作環境

  • git 2.30+ / python3 3.9+(標準ライブラリのみ)/ POSIX sh + bash(macOS 標準の bash 3.2 で動作確認済み)
  • 任意: gh(GitHub CLI)— ある場合のみ done が PR を自動作成します(無くても全機能動作)
  • CI: ubuntu-latest / macos-latest で tests/smoke.sh(29検査)+ tests/breaker.sh(14検査)を実行

導入(リポジトリごとに1回)

./install.sh /path/to/your-repo

push 前に何が入るか確認したい場合は `./install.sh /path/to/your-repo --no-push`(origin への自動 push を保留します。内容を確認したら `--no-push` 無しで再実行すると push されます)。

これで対象リポジトリに以下が入ります。

  • .agents/bin/agents — 調整 CLI(python3 標準ライブラリの