auto-review-loop

Automated iterative research review loops are executed by running a cycle of review, fix, and re-review until a positive assessment is achieved. This skill handles plan parsing, Codex MCP verification, and result documentation autonomously.

Plan-driven mode executes dependency-aware TODO items in topological order, verifying each with Codex MCP. Free-form mode iterates review and fix cycles rapidly without a formal plan, stopping only when a positive assessment is reached.

Yes, Codex MCP and ARIS integration are required to run autonomous review workflows. The system uses Codex MCP for task verification and ARIS integration to coordinate remote runs, track progress, and document results.

Parallel experiments are routed across multiple servers using optional remote-dispatch to GPUs. The workflow manages state persistence and server dispatch to coordinate parallel experiment routing and track progress.

Yes, autonomous review can be used for prompt refinement without a formal plan. The free-form mode iterates the review and fix cycle continuously until the prompt reaches a positive assessment.

Plan-driven workflows require a dependency-aware implementation plan to function. Without a formal plan, the system cannot execute topological task ordering and must fall back to free-form iterative refinement.