orc-challenge

What problem does it solve? Reviewing a finished artifact — a TSD, PRD, ADR, API contract, README, runbook, plan, or code module — is unreliable when the reviewer also wrote or fixed it, because a session that wrote the fix grades its own homework. This Skill separates the judge from the fixer: it grades the artifact against a goal you state, records findings, and stops, forcing fixes to happen in a fresh session. ## Core Features & Use Cases - Deterministic lint plus multi-lens council review: runs a zero-token lint, then dispatches parallel review lenses (cold reader, contrarian, outsider, executor, first-principles thinker, expansionist) whose findings a judge resolves with full conservation tracking. - Computed pass/fail with an auditable ledger: the CLI computes the verdict, enforces 100% finding coverage across iterations, detects tampering, and tracks stalled cycles instead of capping loops. - Fix briefs instead of in-session fixes: on failure it writes a paste-ready fix brief and stops; on pass it produces a final report with opportunities and premise challenges clearly marked as non-blocking. - Use Case: You finished a technical design document and want to know if a backend team can implement it without asking you questions. Run the challenge, answer the intake questions about goal and audience, receive graded findings across seven dimensions, fix in a new session, and resubmit until it passes. ## Quick Start Ask the assistant to run /orc-challenge on your finished design document and answer the seven intake questions about goal, audience, and review council.