What problem does it solve? Issue boards and audit passes only produce incremental fixes, so teams keep hill-climbing the current design without ever questioning its shape. This Skill runs a weekly structural review that reads the board as a symptom, mines the repo's measured evidence, and sweeps outside the repo for lifted platform constraints, producing ranked proposals for changes that remove constraints or eliminate whole classes of work. ## Core Features & Use Cases - Three-layer evidence sweep: Reads cluster labels and recurring issue patterns on the board, internal measured evidence (e2e corpus, latency reports, judge audits, the nothing-checks register, ADRs), and external sources (platform changelogs, current practice, published work) via WebSearch and WebFetch. - Subtraction hunting: Actively searches for mechanisms to retire, guarantees to drop, warn-only checks nobody reads, and lanes to delete, not just additions. - Ledger-backed memory: Carries forward, re-ranks, and records verdicts (accepted, deferred, set aside, rejected) in an ADR ledger so researched-and-rejected ideas are never re-proposed. - Use Case: After the weekly board audit, run the Skill to get a ranked list of structural proposals, each with a pre-verified claim, the constraint it removes, observable outcomes, and a cheapest one-day probe that could kill it. ## Quick Start Run /find-big-wins after the weekly board audit to propose ranked structural bets for the lead to review.
