Kaiju: an agent bake‑off for your repo

Описание и применение

Kaiju позволяет сравнивать несколько кодовых агентов, выполняя один запрос в изолированных ветках, автоматически тестировать результаты и выбирать победителя, после чего объединять изменения в основной ветке.

Задача

Determine which coding agent best solves a task in a specific codebase by running the same prompt across multiple agents and comparing results.

Результат

Kaiju enables side‑by‑side comparison of coding agents, automates testing and ranking, and promotes the best run, providing a reproducible bake‑off workflow.

Рабочий процесс

• Start Kaiju daemon
• Open dashboard or use CLI to submit a prompt and select agents
• Kaiju creates isolated git worktrees and tmux sessions for each agent
• Agents run the prompt and generate code changes
• Dashboard shows live diffs and status
• Press Judge to run a test command (e.g., cargo test) and rank runs
• Select winner and click Promote to merge changes into workspace
• Leaderboard records results for future reference

Доказательства

Kaiju: a small Rust daemon that runs one coding task across several AI CLIs at once (Claude Code, Codex, Gemini), each in its own isolated git worktree, then lets my tests decide the winner and merges that run back into my branch.

Источник

Hit Judge and a local claude -p ranks the runs. Two things keep it honest: The runs are anonymized as A/B/C, so the judge can't favor a brand.

Источник

One click commits the winning run's changes and merges its branch into your workspace. It refuses if your workspace is dirty or the run is still active, and it aborts cleanly on a conflict and tells you which branch to merge by hand.

Источник