Content
63%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 well-structured advisory skill: clear mode detection, a sequenced four-phase workflow, concrete output templates, and deliberate progressive-disclosure design. Its two real weaknesses are that the referenced rules/ and templates/ files are not shipped with the bundle (making the index point at nothing) and that deferral/composability guidance is restated several times.
Suggestions
Ship the referenced bundle files: create rules/ (style-directions, brand-identity, visual-hierarchy, color-systems, typography-pairing, signature-details, style-audit) and templates/direction-brief.md, or inline the essential content — the per-mode loading table currently points at files that do not exist.
State the /ux deferral once (e.g., keep the 'What this skill defers' list, drop the intro restatement and Behavioral Rule 8) to recover token budget in a SKILL.md that positions itself as a thin index.
Add a lightweight verification checkpoint to Phase 4 — e.g., confirm every recommended token actually exists in the project's theme/tokens file before emitting the build sheet.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and instructive with no padding about concepts Claude already knows, but guidance is repeated: the /ux deferral appears four times (intro prose, the Phase 3 table, the six-bullet "What this skill defers" list, Behavioral Rule 8), and the thin-index note appears twice (blockquote plus "Do not pre-load all rules. The progressive-disclosure cost is real."). Could be tightened by stating each once; not 2 because there is no fluff or over-explanation. | 3 / 5 |
Actionability | Concrete, executable guidance throughout: exact mode-detection trigger tokens, platform heuristics ("View, Text, Pressable, expo-router"; "git diff --name-only HEAD~1"), a complete copy-paste build-sheet output template, and a findings format ("cite file:line, the named principle ... a concrete fix with code"). Not 5 because the substantive rules (named directions, palette construction, pairings) are all deferred to referenced files rather than exemplified inline. | 4 / 5 |
Workflow Clarity | A clearly sequenced four-phase workflow (context discovery → load rules by mode → compose with siblings → output) with mode detection up front and explicit decision checkpoints ("If ambiguous, ask"; "Do not review or generate code you have not read"). Not 5 because there is no verification step on the output (e.g., confirming recommended tokens exist in the project's theme); not 3 because checkpoints like ask-when-ambiguous and read-before-review are explicit. | 4 / 5 |
Progressive Disclosure | The body is a well-signaled, one-level-deep index with an excellent per-mode loading table — but scored against the actual bundle, none of the eight referenced files (rules/style-directions.md, rules/brand-identity.md, rules/visual-hierarchy.md, rules/color-systems.md, rules/typography-pairing.md, rules/signature-details.md, rules/style-audit.md, templates/direction-brief.md) exist, so every Phase 2 load fails. Structure and signaling are good, but the disclosure chain is broken as shipped. | 3 / 5 |
Total | 14 / 20 Passed |