Review the interface as one system
A strong interface is not six independent audits stapled together. Review the whole experience, let each
skill own its domain rules, then consolidate the evidence into one prioritized verdict.
This skill owns orchestration only. Accessibility rules belong to
; structure to
; copy to
; type to
; color to
; visual polish and motion to
. Never duplicate or override their rules here.
Core Principles
1. Resolve Scope and Mode First
Infer the screen, flow, feature, or repository scope from the request and current workspace. State the resolved scope in the output. Use
when no mode is supplied.
| Mode | Coverage | Finding cap |
|---|
| Primary user path and highest-traffic states; report only and issues | 5 |
| Entire requested scope across all six domain skills, including empty, loading, error, and narrow-width states when present | 15 |
If the requested scope is too large to inspect credibly, narrow it to the highest-traffic complete flow and state the boundary. Never imply uninspected surfaces were reviewed.
2. Recon Before Judgment
Identify the framework, styling system, component library, design tokens, supported viewports, and available preview or test commands. Follow the project's established Tailwind, plain CSS, CSS-in-JS, token, and component conventions.
3. Use Domain Skills as the Sources of Truth
Load and apply the six owning skills. In
mode, inspect all six domains but spend depth only where the primary flow has evidence. In
mode, complete each domain review before consolidation.
Review in this order so foundational failures are not hidden by polish:
When two skills appear to cover the same issue, assign it to the skill that owns the underlying rule and mention secondary effects in the Why cell. Report it once.
4. Require Evidence
Every finding cites
and shows the current implementation. If the review artifact has no source files, cite the exact screen and component. Do not report a code-level finding from visual appearance alone or a visual finding from source code alone when runtime behavior determines the result.
5. Rank by User Impact
Use one shared severity scale:
- : blocks a task, misleads the user, hides content or controls, causes data-loss risk, or creates a repeated systemic failure.
- : meaningfully harms comprehension, efficiency, adaptability, or consistency.
- : isolated polish with limited task impact. Include only in mode.
Within a severity, rank by reach and leverage. A token or shared-component fix outranks the same symptom in one leaf component.
6. Consolidate Systemic Findings
One root cause is one finding. List every confirmed location in the same row rather than producing a row per occurrence. Do not pad the report to reach the finding cap; a short review or no findings is a valid result.
7. Make Restraint Visible
Record candidates considered but deliberately rejected. A candidate is rejected when the owning skill permits the current implementation, evidence is insufficient, the project convention is intentional, or the proposed change would add complexity without user benefit.
8. Verify What Can Be Verified
Run safe, relevant checks available in the project. Inspect the rendered interface when runtime behavior or visual judgment matters. Report the exact command or interaction and observed result. If a check cannot be run, label it Not verified and state what remains; never convert a verification gap into a finding.
9. Review Without Mutating by Default
Treat a review request as read-only. Do not edit source code unless the user also asks to implement the findings. When implementation is requested, preserve the consolidated report as the change scope and re-run the relevant verification afterward.
Common Mistakes
| Mistake | Fix |
|---|
| Six disconnected domain reports | Consolidate into one ranked findings table |
| Same issue reported by multiple skills | Assign it to the skill that owns the underlying rule |
| Finding with no exact location | Cite and the current implementation |
| Visual claim inferred only from source | Inspect the rendered state or mark it not verified |
| Unlimited low-impact polish | Respect the mode cap; omit findings in |
| Silent gaps in coverage | Show which domains and states were actually inspected |
| No rejected candidates | Include the required considered-but-rejected table |
| Review silently edits code | Stay read-only unless implementation was requested |
| “Approve” with pending actionable findings | Use or |
Review Output Format
Always use the following sections.
Scope and Coverage
State the mode, exact scope, stack and styling conventions, and any review boundary. Then show coverage:
| Domain | Evidence inspected | Result |
|---|
| Accessibility | Files, components, states, or checks | Findings count or |
Include all six domains.
means inspected with no actionable finding;
must explain why.
Findings
Use one table ordered by severity, then reach and leverage:
| # | Severity | Domain | Location | Before | After | Why |
|---|
| 1 | HIGH | Accessibility | | <button><XIcon /></button>
| Add and hide the icon from the accessibility tree | The icon-only control has no accessible name |
Each row is one root cause. The
Domain value is the owning skill without the
prefix. Respect the mode's finding cap. If there are no findings, omit the table and state "No actionable interface findings."
Considered but Rejected
Include 1–3 candidates in
mode and 2–5 in
mode:
| Location | Candidate | Rejected because |
|---|
| Increase the shadow | Existing depth matches the shared surface token; changing one card would reduce consistency |
These are real candidates inspected during the review, not invented filler. If the scope genuinely contains fewer borderline candidates, include the ones that exist and say so.
Verification
List each check or interaction, the exact command or steps, and the observed result. Separate checks that passed from checks marked Not verified.
Verdict
End with exactly one:
- — one or more findings remain.
- — only or findings remain.
- — no actionable findings remain and the claimed coverage was verified.