What problem does it solve? Runtime SKILL.md body content in mj-agent is literal LLM context, so editing it directly risks silent behavioral failures like wrong answers or hallucinations. This Skill enforces a read-only, human-in-the-loop review process that audits, drafts, and impact-analyzes changes before any write occurs. ## Core Features & Use Cases - Five-Section Body Audit: Evaluates SKILL.md bodies against the Agent_Side five-part structure (Purpose, When to use, Planning workflow, Common patterns, Anti-patterns) with quality ratings. - Reverse Reference Scan: Greps agent.py, system.md, docs/, and sibling SKILL.md files to find cross-references that must stay in sync with proposed changes. - Draft-Only Diff Proposals: Produces unified diffs, frontmatter version bump suggestions, impact analysis, and structured HITL questions for Domain Expert and Prompt Engineer review. - Use Case: When a user asks to upgrade the query-writing skill to the five-section format, this Skill audits the current body, scans references, and outputs a proposed diff that pauses for human approval before /mj-agent-doc-author writes it. ## Quick Start Ask the assistant to improve the runtime SKILL.md for a specific mj-agent skill, for example by saying "improve runtime SKILL query-writing", and review the proposed diff it returns.
