What problem does it solve? Getting a second opinion on a plan, architecture, or release candidate from an external model often produces unverifiable claims, lost raw responses, and uncontrolled edits. This Skill structures the entire critic workflow so every finding is preserved, source-verified, and safely incorporated. ## Core Features & Use Cases - Pinned-model critic runs: Query the provider's live catalog, pin exact model IDs, and preflight credentials without exposing them. - Full provenance capture: Save the exact prompt, raw response, model IDs, finish reason, token usage, cost, and checksums for every artifact. - Source-verified findings: Classify each blocking finding as ACCEPT, PARTIALLY ACCEPT, or REJECT with file/line evidence before amending anything. - Safe amendment workflow: Preserve the original artifact, add a revision-basis note, and keep every commit boundary green. - Use Case: Before implementing a large architecture plan, send it to a pinned Claude or Gemini model via OpenRouter, verify each critique against the repository, then amend the plan with only the validated findings. ## Quick Start Ask the assistant to have a named external model review your implementation plan or release candidate and verify the findings before revising it.
