Content
85%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.
A highly actionable, well-structured search-tool skill: every mode has executable commands, the workflow has explicit verification and fallback behavior, and the bundle files are real, well-signaled, one level deep, with accurate section and rule-ID pointers. The main cost is redundancy — the data inventory is restated from the frontmatter, and the retry/query-shaping rules are repeated across three sections.
Suggestions
Deduplicate the retry and fallback guidance: it currently appears in the Query Contract ("Retry once with a narrower rewrite..."), in "If a search returns 0 results" ("Retry once with a narrower query or an explicit domain/stack"), and again in "Tips for Better Results" — keep one canonical statement and cross-reference it.
Delete the opening "Searchable local UI/UX guidance: 79 searchable styles (50 active), 192 product palettes..." sentence from the body: it repeats the frontmatter description almost verbatim and spends tokens restating what the user already matched on, rather than instructing.
Merge the "## Workflow" heading with the "## Query Contract" section that immediately follows it (two back-to-back H2s where the contract is effectively the workflow's preamble), and fold the accessibility and text-layout query examples into a single "semantic outcome first, then stack" pattern to avoid two near-parallel paragraph blocks.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is largely operational (commands, tables, dials), but it repeats itself: the frontmatter's data inventory is restated almost verbatim in the opening line ("Searchable local UI/UX guidance: 79 searchable styles..."), the "retry once with a narrower query" rule appears in the Query Contract, again in "If a search returns 0 results", and again in "Tips for Better Results", and the "one dominant intent / 2–5 meaningful terms" rule appears in both the Query Contract and Tips. Not 4 because the duplication spans multiple sections rather than being minor over-explanation; not 2 because there is no padding that explains concepts Claude already knows — everything is skill-specific. | 3 / 5 |
Actionability | Fully executable, copy-paste-ready commands throughout: every mode (design-system, domain, stack) has a complete invocations with the full script path, concrete example queries for each domain (e.g. "decorative icon aria hidden" --domain icons), concrete dial values with behaviors, and an end-to-end example workflow. The only placeholder inputs are the user's own query and project name, which is inherent to the tool. | 5 / 5 |
Workflow Clarity | The workflow is clearly sequenced (Step 1 → 2 → 2b → 2c → 3 → 4) with explicit validation and feedback loops: verify "the returned domain/category, top result identity, and fit" before applying results, retry-once-then-fallback-on-empty handling with instructions never to present 0-result output as data, MASTER.md existence checks before --persist, and a canonical pre-delivery checklist in references/pro-rules.md. All three elements of the top anchor (validation steps, error-recovery loops, checklists) are present. | 5 / 5 |
Progressive Disclosure | The body is an overview that explicitly defers detail to real, one-level-deep, well-signaled references: "The full rule text for every category lives in references/quick-reference.md — read it on demand rather than loading it every time", plus references/pro-rules.md for the pre-delivery checklist. Both files exist, the §3/§5/§6/§7/§8/§9 section pointers in the Tips table match the actual section headers, and the cited rule IDs (color-dark-mode, spring-physics, inline-validation, nav-hierarchy, mobile-first, virtualize-lists, main-thread-budget) all exist in quick-reference.md. The inline priority table is a compact summary with the full list correctly pushed to the reference file. | 5 / 5 |
Total | 18 / 20 Passed |