omk-kimi-runtime

What problem does it solve?

This Skill prevents unreliable agent behavior by standardizing Kimi runtime choices, tool usage, context loading, and quality gates for oh-my-kimi workflows.

Core Features & Use Cases

• Mode Selection Guardrails: Uses thinking mode for architecture, coding, debugging, review, planning, and multi-step tool work, and uses no-thinking mode for fast summarization, commit messages, simple classification, and web-heavy research.
• Tool Use Discipline: Keeps tool calls scoped to the current task, inspects nearby code and conventions before writing files, and prioritizes multimodal evidence (screenshots/traces/video) before proposing changes.
• Okabe + D-Mail Context Insurance: Ensures generated agents inherit the Okabe-compatible base (enabling SendDMail), using D-Mail for checkpoint rollback scenarios like risky refactors, migrations, context compaction, or long-horizon work that could lose state.
• Context Policy for Large Repos: Builds a file map first, reads only relevant files, relies on project-local graph memory, and checkpoints/summarizes to avoid context pressure.
• Completion Rule (Quality Gates): Declares a task incomplete until lint, typecheck, tests, optional build, and diff review pass.

Quick Start

Use the omk-kimi-runtime skill to run your oh-my-kimi workflow on Kimi K2.6 with thinking/no-thinking mode selection, scoped tools, safe context loading, and enforced quality gates.