Studio Audit

studio-audit

Use this when the user feels "done" and wants the work judged like a senior designer would judge it before handoff. This is an umbrella review skill, not a new visual system. It composes the existing Cortex design review and craft skills, then synthesizes one priority-ordered report.

Its domain-owner coverage and evidence discipline incorporate Jakub Krehel's better-interface at commit a67333399dabbc71d7778962cb9c4fb9b86a00d0.

law

Read references/studio-law.md before judging. Read references/report-template.md before writing the final report.

Default to audit-only. Do not edit code, run formatters, or apply fixes unless the user explicitly asks for fixes. If a delegated skill offers to fix safe small issues, keep it in report-only mode unless the invocation includes fix, apply, or equivalent explicit permission.

modes

Default to full. If the requested scope is too large to inspect credibly, narrow it to the highest-traffic complete flow and state the boundary. Never claim coverage for an uninspected surface.

workflow

1. resolve scope and mode. Identify whether the target is a component, page, flow, whole app, screenshot, URL, or branch. If scope is missing, audit the current visible app or branch and state that assumption. This step is complete when the mode, exact target, and review boundary are explicit.

2. load project truth. Read local design docs, existing component conventions, recent changed files, and any asbuilt package if one is present. Do not require asbuilt; treat it as extra project-specific evidence. This step is complete when the stack, styling system, behavior engine, tokens, supported viewports, and available preview or test commands are accounted for.

3. review every domain. Load the owning skills in this order:

   • better-accessibility, with preflight for broad static risks
   • better-layout, with responsive-craft for viewport behavior
   • better-writing
   • better-typography
   • oklch-skill
   • better-ui

   When orchestrated here, use their principles and references but ignore their standalone output formats. In quick mode inspect every domain and spend depth where evidence shows the greatest risk. In full mode complete each domain review. Mark any unavailable or untested domain not covered; never reconstruct its rules from memory.

4. inspect the rendered product. Use responsive-craft preview or its snapshot workflow when the target can run in a browser. Cover mobile and desktop at minimum. Use wip-senior-audit for live first impressions, navigation, hierarchy, states, trust, and conversion. A claim that depends on runtime behavior or appearance requires rendered evidence; otherwise mark it not verified.

5. apply Studio taste. Use interface-craft critique for composition, hierarchy, density, user context, and cohesion. Use emil-design-eng review for motion, feedback, timing, interruption, and interaction feel. Its standalone verdict yields to this audit's one final verdict. Studio law and the project profile adjudicate every supplier opinion.

6. synthesize one verdict. One root cause is one finding even when several domains observe it. Cite a file and line, route, viewport, screenshot, or interaction step for every finding. Rank by Studio priority, then reach and leverage; a shared token or component outranks the same symptom in one leaf. Record real candidates considered and rejected because the evidence, project convention, or user value did not justify a change. The audit is complete when every domain has a coverage result, every reported finding has evidence and a concrete fix, and every verification gap is named.

output

Write the final answer in the format from references/report-template.md.

If working inside a repo and the audit is more than a quick chat critique, also write the durable report to docs/design-audit/studio-audit.md. If screenshots were captured, store them under docs/design-audit/assets/ and reference them from the report.

Close with:

• ship: no blocking issues
• fix first: P0/P1 issues exist
• review again: the audit could not cover the live flow, mobile, or another required surface