kodo

About

kodo exists to convert idle-night time and flat-rate agent subscriptions into finished, verified work: an orchestrator LLM delegates tasks to coding agents through work cycles with role separation (workers, architect verifier, testers), optionally parallelized across git worktrees. Modes include goal-building runs, kodo test for user-style bug hunting, and kodo --improve for senior-developer-style review with --fix-from to repair earlier findings. Because agents run with bypassPermissions, the docs direct users to commit or back up first. Individual developers on Claude Code Max subscriptions use it for unattended feature work and codebase improvement.