Content
78%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
An exceptionally dense, opinionated, and executable rulebook: almost nothing is background knowledge Claude already has, and the verification procedure is concrete. The weaknesses are structural repetition across the density-related sections and a 335-line monolithic body that should offload the shadcn and agent-surface rulebooks to reference files the way it already offloads visual direction.
Suggestions
Move the shadcn/ui Design Rules and Agent Surface And Page Boundary sections into a references/ file (e.g., references/shadcn-rules.md and references/agent-surface.md), keeping a 2-3 line summary plus a clearly signaled link in SKILL.md, as is already done with references/visual-direction.md.
Deduplicate the density rules: the 'title or description, never both', 'no more than four visible choices', and nested-cards rules each appear in 2-3 sections (first-viewport budget, Default Surface Density, Product Surface Review, Anti-Patterns) — state each rule once and cross-reference it.
Make the implementation throughline explicit: after the one-line contract step, add a short ordered sequence (pick direction -> build smallest cohesive surface -> run Product Surface Review checklist -> run Verification) so the thematic sections read as stages of one workflow.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Nearly every token is product-specific knowledge Claude cannot infer (repo easing tokens like `var(--ease-collapse)`, `guard:no-default-chrome`, `sendToAgentChat` contracts, Tabler icon verification), with no filler explanations of concepts Claude already knows. It misses anchor 5 because rules repeat across sections: the "title or description, never both" and four-visible-choices rules each appear in roughly three places (first-viewport budget, Default Surface Density, Product Surface Review, Anti-Patterns). | 4 / 5 |
Actionability | Guidance is copy-paste executable throughout: `pnpm dlx shadcn@latest add <component>`, grepping `node_modules/@tabler/icons-react/dist/tabler-icons-react.d.ts` for an exact icon name, token names with timings (`var(--ease-drawer)` 260ms), `ease-[var(--ease-collapse)]` in Tailwind, and concrete primitive choices (Switch for binary, ToggleGroup for small option sets). This matches anchor 5: specific examples cover the common cases. | 5 / 5 |
Workflow Clarity | There is a clear pre-coding checkpoint (write the `<mode> / <user> / <job> / <next action>` contract before implementing) and a numbered 6-step verification procedure with escalation criteria, state checks (hover, focus, loading, empty, error), and a conformance gate for adapter changes. It stops short of anchor 5 because the implementation phase between contract and verification is organized thematically rather than as one coherent sequence, so a reader assembles the throughline from separate sections. | 4 / 5 |
Progressive Disclosure | The one bundle reference (`references/visual-direction.md`) is real, well-signaled, and exactly one level deep, cited both in the Visual Direction Contract and Verification sections. But the 335-line body inlines entire rulebooks that belong in separate reference files (shadcn/ui design rules, agent surface and page boundary rules, anti-patterns), and the Related Skills list confirms sibling skills exist, so this matches anchor 3 (content that should be separate is inline) rather than anchor 4. | 3 / 5 |
Total | 16 / 20 Passed |