autoresearch-qa-skills 0.3.2

Machine-verifiable dual-AI release QA: clean-session ZIP-only provenance, immutable per-review reports, fail-closed consensus, source-bound runtime checks, release authoring evidence gates, and deterministic one-skill-per-ZIP artifacts.\n\nThis release intentionally does not claim a quantified accuracy gain; routing and dual-review efficacy require retained model traces and benchmark labels.