Content
67%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.
The body is a well-structured overview with genuinely specific design guardrails and a clear entry-to-exit flow that delegates detail to annotated one-level references. Its weaknesses are a large block of scheduling boilerplate that inflates token cost without adding actionable guidance, and a heavy reliance on reference files that are not present in the bundle, which makes the checklist and phase model unverifiable.
Suggestions
Collapse or remove the Scheduling metadata block (intent signature, expected inputs/outputs, dependencies, control-flow features) — its content is already covered by When to use, Guardrails, and References, and it costs roughly a third of the body's tokens.
Inline the checklist's key mechanical checks (or at least name them) instead of referencing 'Phase 6 (checklist section 6 mechanical checks)', which points to phases and sections not defined in this file.
Ship the referenced files with the skill or inline the most critical ones (anti-patterns, checklist) — none of the 27 referenced resources/ or reference/ paths exist in the bundle, so the core workflow cannot actually be executed from what is provided.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Line-by-line the writing is terse and avoids explaining concepts Claude already knows, but the ~55-line 'Scheduling' block (intent signature, expected inputs/outputs, dependencies, control-flow features, resource scope, preconditions) is orchestration metadata whose content largely duplicates the When-to-use and Guardrails sections. This fits 'mostly efficient but includes some unnecessary explanation or could be tightened' rather than 2, since no section is padded prose or a tutorial on known concepts. | 3 / 5 |
Actionability | Guardrails give executable specifics: the exact system font stack ("system-ui, -apple-system, BlinkMacSystemFont, 'Segoe UI', Roboto, sans-serif"), a ranked CJK font order ("Pretendard Variable > Noto Sans CJK > system-ui fallback"), a concrete naming format ("Deep Ocean Navy (#1a2332)"), a runnable command ("bunx getdesign@latest list"), and hard rules ("Div-based fake screenshots are banned"). It is not 5 because the canonical workflow path itself is abstract ('Generate or revise DESIGN.md, then run the design checklist') and delegates key operational detail entirely to external files. | 4 / 5 |
Workflow Clarity | Entry, Transitions, Failure and recovery, and Exit give a clear sequence with checkpoints ('If accessibility checks fail, revise before handoff', 'audit against them in Phase 6'). It is not 5 because 'Phase 6 (checklist section 6 mechanical checks)' references phases and checklist sections not defined in this body, leaving the validation loop implicit and delegated to execution-protocol.md and checklist.md; it is above 3 since validation and error-recovery branches are explicitly present. | 4 / 5 |
Progressive Disclosure | The References section lists 27 files, each annotated with its purpose (e.g. "resources/redesign-protocol.md: existing-site redesign"), one level deep and clearly signaled, matching 'good structure; most content appropriately placed'. It is not 5 because none of the referenced paths exist in the provided bundle (no references/, scripts/, or assets/ directories are present, and the referenced resources/ and reference/ directories cannot be verified), and the near-identical directory names 'resources/' and 'reference/' blur navigation. | 4 / 5 |
Total | 15 / 20 Passed |