What problem does it solve? Research work needs rigorous, adversarial review before submission, but manual peer review is slow and self-review is biased. This Skill runs an autonomous review-fix-re-review loop with an independent reviewer model until the work earns a policy-approved positive assessment or the round limit is reached. ## Core Features & Use Cases - Autonomous Review Loop: Iterates review, fix implementation, and re-review for up to 4 rounds, stopping only when score >= 6/10 and verdict is "ready" or "almost". - Multi-Backend Reviewer Routing: Selects reviewers across Copilot native rubber-duck subagents, Codex MCP, or manual review, enforcing opposite model families for independent assessment. - State Persistence & Recovery: Writes REVIEW_STATE.json and an append-only ACQUITTAL_LOG.jsonl so loops survive context compaction and remain auditable per run_id. - Adjustable Difficulty: Medium, hard (reviewer memory + debate), and nightmare (direct repo access + adversarial verification) modes control how adversarial the reviewer is. - Use Case: After finishing experiments for a paper draft, run the loop so an independent GPT-family reviewer attacks your Claude-written claims, you implement the minimum fixes, and the cycle repeats until the work is judged submission-ready. ## Quick Start Ask the agent to run the auto review loop on your current research draft, for example: "Run auto-review-loop on the paper draft and experiment results in this project."
