What problem does it solve? Sub-agent briefs written from instinct repeat the same defects: agents hang in wait loops, lose uncommitted work, game metrics, or report success without evidence. This Skill turns each paid-for failure into a required clause and a linter rule, so every new brief starts from the last one's lessons instead of from memory. ## Core Features & Use Cases - Defect register: A table of real prompt defects (hung agents, lost trees, gamed metrics, unverified evaluator findings) with the exact clause that prevents each one. - Brief structure checklist: Eight required parts covering scope, ground truth, operational metric definitions, definition of done, anti-gaming, commit discipline, and the evidence contract. - lint-brief.mjs: A Node.js linter that fails a brief missing any load-bearing clause, flags metric words used without numbers, and ships with a --selftest suite. - Use Case: Before delegating a UI rebuild ticket to a build agent, write the brief, run the linter, fix the flagged clauses, then spawn the agent with a brief that cannot be misread. ## Quick Start Write your sub-agent brief to a text file and run node .claude/skills/subagent-briefing/lint-brief.mjs on it, then fix every flagged clause before spawning the agent.
