opus-mind

Use when the user edits CLAUDE.md / AGENTS.md / .cursorrules / GEMINI.md / **/SKILL.md or any chatbot system prompt (LINT), OR when the user wants to tighten a vague one-shot prompt before sending it to an LLM (BOOST). Fires on audit/score/review/fix requests, on symptoms like refuse-relent, narration leak, rule conflict, and on "help me improve this prompt" messages.