oh-my-kimi — Evidence-gated Agent Runtime for Kimi

Main

• Use evidence gates to stop “done by narration”: require a diff, a passing command, or a file existence check before accepting completion.

• Run parallel work in worktrees to avoid context corruption; merge only after you inspect evidence and summaries.

• Treat .omk/ as your run ledger: keep it in your workflow so runs are replayable, auditable, and shareable when you debug reliability.

Source-backed notes

• README presents OMK as a Kimi-native control plane with DAG-based execution, evidence-gated completion, and worktree isolation.
• README quickstart shows npm install -g @oh-my-kimi/cli then omk init, omk doctor, omk chat.
• README references verification commands like omk verify --json and operator UIs like omk cockpit/omk hud for visibility.

FAQ

• Is OMK a prompt pack?: No — README explicitly says it’s a runtime/control plane, not a prompt pack.
• Do I need to learn DAGs to start?: No — start with omk chat; use DAG/parallel features once you trust the basics.
• What’s the minimal verification loop?: Run omk doctor, execute a small task, then check omk verify --json before calling it done.