What problem does it solve? Manually cycling between code review and fixes is slow and error-prone, especially when tracking findings across multiple iterations. This Skill automates the review-remediate loop: it dispatches a code-refiner subagent that alternates between reviewing a git diff and fixing flagged issues until the change is approved or the iteration budget runs out. ## Core Features & Use Cases - Automated review-remediate loop: Dispatches code-reviewer and coder subagents across configurable model tiers, iterating until approval or budget exhaustion. - Provenance-validated review artifacts: Every review file carries a reviewer provenance line validated against allowed model tiers, preventing silently degraded or improvised reviews. - Era-based versioned reviews and carry-over: Review files are versioned per era (e.g., review-v1.md, review-v2.md), and a prior era's findings can be carried over for a targeted remediation pass. - Use Case: After executing an implementation plan, run refine-code with the pre- and post-implementation SHAs to get a cross-provider verified review, automatic fixes for Critical/Important findings, and a structured STATUS summary consumable by execute-plan. ## Quick Start Ask the AI to run the refine-code skill on the current git range, providing the base SHA, head SHA, and a description of what was implemented.
