Automates detection and prioritization of audit-related issues across the workspace. Scans for performance, accessibility, and monitoring signals. Use when analyzing audit errors or requesting actionable improvement patches.
インストール
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
Automates detection and prioritization of audit-related issues across the workspace. Scans for performance, accessibility, and monitoring signals. Use when analyzing audit errors or requesting actionable improvement patches.
Audit Analyzer
Context
Automates detection and prioritization of audit-related issues across the workspace. Scans documentation and code for performance, accessibility, and monitoring signals; outputs actionable improvements and small, safe patches. Triggers on phrases like "Analyze audit errors" or "Audit analysis".
Instructions
Discover audit signals across the workspace
Scan for common audit signals: Lighthouse performance hints (LCP, FID, CLS), Lighthouse scores, caching headers, accessibility hints (aria-labels, alt text), error monitoring (Sentry, uptime), and common deployment concerns (CI/CD, rollback, smoke tests).
Identify files and sections referencing these signals.
Classify and prioritize improvements
Group findings into categories: Performance, Accessibility, Monitoring/Errors, Build/Deployment.
Score items by impact vs effort to get 80/20 high-leverage fixes.
Propose concrete patches (small, safe changes)
For each high-priority item, provide a minimal patch snippet and the target file path.
Include a short rationale and a quick test step.
Output format
Return a compact plan with: Category, File, Patch Snippet, Why it helps, and Test steps.
Apply or stage changes
Suggest applying changes directly or creating a patch file for review.
Do not push to remote; leave review to user.
Optional: Create a new SKILL.md on success
If user approves, codify this as a new skill under /workspace/skills/audit-analyzer/SKILL.md
Constraints
Do not perform destructive operations without explicit user approval.
Focus on small, incremental changes that are safe and testable.
Do not modify tests unless explicitly requested.
Output only actionable patches and plan summaries.