What problem does it solve? After running a convention audit, teams face dozens of low-to-medium severity code convention deviations (missing scoped styles, missing cid attributes, import order issues) that are tedious to fix by hand but risky to change blindly in bulk. This Skill applies controlled, user-confirmed fixes to those reported issues without touching business logic. ## Core Features & Use Cases - Controlled issue fixing: Reads the convention-audit report, resolves each issue's file, line, and deviation type, then applies rule-based or AI-based patches only after a diff preview is confirmed. - Severity gating: Fixes only medium and low severity deviations; critical structural issues are refused and routed to page regeneration or manual work. - Mandatory re-validation: After each fix round it automatically re-runs page validation and reports remaining errors, closing the loop before commit. - Use Case: Before a pull request, run the audit, then ask to fix all rule-based deviations in the report; review each diff group, confirm, and get a re-scan summary showing zero remaining errors. ## Quick Start Ask the AI to fix all medium and low severity issues listed in reports/规范审查报告.md with diff previews before each change.