auto-review-loop

An autonomous research review loop automates iterative cycles of assessment, fix implementation, and re-review until a positive verdict or maximum round threshold is reached. It enforces a structured Phase A-E workflow with detailed logging to persistent documents.

State persistence for long-running review workflows writes a JSON state file after each round, capturing progress and enabling recovery. This supports compact recovery and allows resuming the review process directly from previously completed rounds.

Yes, the automated multi-round review process supports optional human checkpoints. Enabling this flag pauses the autonomous workflow at defined phases, allowing manual oversight before continuing with fix implementation and re-review cycles.

The maximum number of review rounds is defined by a configurable MAX_ROUNDS constant. The autonomous loop continuously cycles through review and fix implementation until either reaching this limit or achieving a positive assessment threshold.

To resume an interrupted autonomous review loop, the workflow utilizes compact recovery from the persistent JSON state file. This recovery mechanism reads the saved state to bypass completed phases and continue the review process from the exact interruption point.

You should avoid an autonomous multi-round review loop when research artifacts require subjective human judgment beyond optional checkpoints, or when iterative automated fix implementation cannot adequately address the specific assessment criteria of your workflow.