Cuts a prompt to its minimum effective length — same behavior, fewer tokens, receipts for every cut.
Makes the model ask the 3-5 questions that would most change its answer before it answers.
Critiques a prompt against lab-consensus rules and returns a rewrite plus a numeric scorecard — not vague praise.
Produces a one-page brief that gets a new developer productive in the codebase on day one.
Audits third-party dependencies for risk: maintenance health, blast radius, replaceability, and lock-in.
Ranks technical debt by interest paid, not by ugliness — what actually slows the team down.
Builds a structural map of an unfamiliar codebase: entry points, core modules, data flow, and the 20% that matters.