Agent
Accessibility audit of an implemented UI surface — assistive-tech labels, hit-target size, contrast ratios, text scaling, reduced-motion. Read-only; produces an S/A/B/C/D/F per-dimension report where a missing label or sub-threshold hit target blocks ship regardless of visual grade. Run after a screen is…
Agent
Token-discipline sweep — greps source for raw color literals (hex / rgba / rgb / hsl, inline-style color props, Tailwind arbitrary color values) that bypass the design system, classifies each by severity (S0 chrome / S1 content / S2 internal / Exempt), and proposes the closest semantic token. Cheap, periodic…
Agent
Multi-screen arc audit — grades CONTINUITY in a multi-screen user-flow journey (onboarding, checkout, setup, sign-up→first-use, deletion) via two input modes — (a) walks a live flow end-to-end and grades eight gap classes that single-screen reviewers structurally miss, writing a canonical flow doc plus a dated…
Agent
Semantic chrome-vs-handler integrity audit of a single screen — builds a per-element affordance-vs-behavior table (what the chrome PROMISES vs what the handler DOES), then flags dead chrome, redundant affordances, and optical-group disconnects. Read-only; produces an S/A/B/C/D/F report. Run after a screen is…
Agent
Cross-section consistency auditor — compares the project's primary multi-section surface (tab bar / dashboard sections / top-level views) side-by-side to find consistency drift no single-screen review catches: header pattern, section-title typography, card/row pattern, empty state, CTA hit-size, surface treatment.…
Agent
Senior-IC product designer for information-architecture / user-flow / multi-screen work — the gateway between brainstorm and implementation. Produces a spec doc (IA + flow + per-screen state inventory + considered-and-rejected + engine asks) for new features and redesigns; the spec IS the deliverable, not a mockup.…
Agent
Single-screen visual-polish audit — grades a captured screen against the project's own quality bar (or platform-native conventions if none), names a chrome reference and a domain reference, and scans five composition pitfalls. Read-only; produces an S/A/B/C/D/F report anchored to named references plus one highest-ROI…
Skill Claude CodeCodex
Elicit the project's design north-star + design-system reference (the two artifacts the consumed design audits read at runtime); no agent copies are authored. Derived from the project's actual UI surfaces, the user's named design benchmarks, and past design failure modes. Invoke /dotclaude:design in any project root…
Skill Claude CodeCodex
Before reusing an existing string, component, or copy pattern on a NEW surface — locate its current usage with grep, classify both the existing and the proposed surface by type, and apply the verdict matrix. Catches first-touch copy leaking onto a daily-driver surface, a bug invisible to lint, code review, and…
Skill Claude CodeCodex
Drive a multi-screen flow from raw to award-quality through a bounded iterative loop — reset to a clean fixture, capture every state, grade across three scrutiny layers (reviewer agent + composition scan + backend-truth probe), apply ONE highest-ROI fix, commit atomically, re-run — until award-quality OR a hard…
Skill Claude CodeCodex
Build the prior-surface inventory before designing OR auditing any screen — walk every surface the user touched before reaching this one, classify each by type, and apply the forbidden-pattern matrix so first-touch copy/voice/chrome never leaks onto a daily-driver surface. Load before any new design spec or any…
Skill Claude CodeCodex
Run three orthogonal outside-eyes tests on every user-visible string — day-30 (frequency-jaded), partner (voice-register), and stranger (already-knows-me) — to catch voice violations that pass a deny-list but still feel wrong. Load at design time on every proposed string and at audit time on every visible string on a…