What problem does it solve? High-stakes changes like implementation diffs, release closeouts, and plans need independent review evidence, but a single model review can miss defects. This Skill convenes a multi-vendor advisor board through the agent-harness runtime so each artifact is reviewed by several independent model seats with distinct lenses. ## Core Features & Use Cases - Cross-vendor review board: Composes a default 4-seat board (Claude Fable 5, Grok 4.6, GPT-6 Astra, Gemini 3.8 Flash) with distinct review lenses, availability-aware backfilling, and a hard floor of 3 reviewers. - Flexible material input: Feed artifacts inline, stage local files via artifact_ref/brief_ref, or pass large private files by reference with context_refs so legs open them with their own tools. - Bounded review loops: Heartbeat-based liveness detection, input-scaled timeouts, delta re-review of dissenting seats, and round caps that converge a review-fix-re-review loop. - Use Case: Before merging a risky refactor, run phase-loop advisor-board <artifact> to collect AGREE/DISAGREE verdicts from four independent vendors, then re-run only dissenting seats against the delta after each fix. ## Quick Start Ask the agent to run the advisor board on your change artifact, for example: run a cross-vendor advisor board review of my implementation diff at path/to/bundle.md and report each seat's verdict.
