What problem does it solve? Asking a model to write language best-practice rules directly produces plausible, uncited advice that changes no behavior, and researching only a handed-down topic list misses the topics nobody knew to ask for. This Skill runs a structured, self-directed research program that grounds findings in real codebase measurements, discovers topics from primary corpora, and ends in publishable AI-config rules and skills. ## Core Features & Use Cases - Eight-phase research loop: Frame, Ground, Scout, Map, Dive, Consolidate, Iterate, Author, and Validate, repeating until a wave produces no new MUST rules. - Worker prompt contracts: Verbatim output contracts for grounding auditors, scouts, deep-dive researchers, consolidators, and revisers so every artifact merges cleanly. - Artifact validation: A Python script checks name parity, context budgets, link resolution, dead globs, rule-table completeness, and runnable verification commands before publishing. - Use Case: A team wants its coding agents to become expert in Rust. The program audits their codebases, surveys canonical guides, linter catalogs, and failure corpora, consolidates cited rulesets with stable IDs, and ships glob-scoped rules plus a review skill. ## Quick Start Ask the agent to run a research program that turns our team's Rust expertise into cited rules and skills for our coding agents.
