What problem does it solve? Agent-generated answers, skills, tools, and memory edits need consistent quality control before reaching users, but ad-hoc review is subjective and either too lenient or wastefully strict. This Skill defines a supervisor role that judges every draft against explicit criteria and decides whether to approve, revise, or escalate. ## Core Features & Use Cases - Structured scorecard: Scores each draft on pertinence, methodology, objective, brand conformity, and confidence, then issues an approve, revise, or escalate verdict with an actionable correction. - Proportionality rules: Distinguishes real defects (ungrounded claims, wrong question answered, fabricated sources) from style nitpicks, so costly re-runs only happen when justified. - Brand and standards enforcement: Checks artifacts against LaRuche rules such as English code, French brand lexicon, no em dashes, variabilized UI strings, and secret handling. - Use Case: A forager agent drafts an answer claiming tests pass without running them; the supervisor returns a revise verdict with the instruction to actually run the test suite and report the real result. ## Quick Start Review this draft answer against the LaReine charter and return a scorecard with a verdict and correction instruction.
