What problem does it solve? Substantial design and architecture decisions made with AI assistance often go unrecorded or get approved without real human review, leaving teams with no traceable rationale for schema, API, dependency, or boundary changes. ## Core Features & Use Cases - Artifact Routing: A two-gate router decides whether a change needs a governance artifact at all, then selects the right one — ADR (Nygard or MADR), RFD, or design doc — with an exemption path for refactors and trivial fixes. - Weight-Scaled Critic Panel: Heavy decisions get four parallel adversarial critics (alternatives, reversibility, blast-radius, assumptions) plus a synthesizer, looped up to three iterations; light decisions get a single quick review. - Mandatory Human Sign-Off Gate: No artifact is written to docs/ until a human explicitly approves; the decision is recorded with an AI-provenance fence (agent, model, trigger) plus the approver's signature. - Use Case: When deciding "should we use Redis or Postgres for session storage", run the skill to route the decision to an ADR, have the critic panel stress-test the trade-offs, sign off yourself, and land a numbered record in docs/decisions/ linked to the commit and PR. ## Quick Start Ask the AI to govern the decision of whether to migrate session storage to Redis, and approve the drafted ADR when it is presented for sign-off.
