run-repair-loop

The skill presents a coherent, tightly-scoped approach to iterative review-test-fix cycles with explicit boundaries and safeguards. Its footprint aligns with the stated purpose of converging a repository to a clean state while limiting risk through pre-flight constraints and explicit stop conditions. While there is a theoretical risk of command injection if inputs allow arbitrary shell commands, the current design emphasizes controlled test commands and minimal patches, with approval gates for risky actions. Overall, the risk is low-to-moderate and the capability appears proportionate to the described use case.