Content
56%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 router-style SKILL.md with genuinely good routing design — specific symptom-to-file mapping, conflict resolution, and anti-rationalization guidance — but it is padded by four redundant renditions of the same routing map, and every reference it routes to is missing from the bundle, so the progressive-disclosure structure exists only on paper.
Suggestions
Collapse the redundant routing renditions: keep the Quick Reference table and the numbered Decision Tree, and delete the dot digraph, the Critical Patterns section, and the Example Invocations (or keep just one of the latter two) — this would cut the body roughly in half.
Bundle the referenced skills/*.md files (or fix the paths to wherever they actually live relative to this SKILL.md) so the one-level-deep references resolve; a router whose targets are absent provides no navigable content.
Trim the Anti-Rationalization table to the highest-frequency misconceptions (e.g., semantic colors, Liquid Glass ≠ blur, Dynamic Type) and drop rows whose 'Reality' column merely restates the routing table.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The same routing map is restated four times — Quick Reference table (17 rows), dot digraph (~17 edges), numbered Decision Tree list (14 items), and Example Invocations (10 mappings) — plus a fifth restatement in Critical Patterns, which summarizes what each referenced file contains. This is 'noticeably verbose; several unnecessary... padded sections' (anchor 2): the dot graph in particular duplicates the numbered list immediately below it, and while nothing explains concepts Claude already knows, well over half the body is redundant routing. | 2 / 5 |
Actionability | For a router skill, guidance is concrete and executable: each symptom maps to a specific file ('Liquid Glass effects, adoption, migration from blur effects | See skills/liquid-glass.md'), with explicit multi-skill chains ('Read skills/hig.md, then /skill axiom-accessibility') and a Conflict Resolution section giving precedence rules. Minor gaps keep it below 5: a few rows defer vaguely ('See axiom-swiftui') and the auditor row buries its procedure in a parenthetical. | 4 / 5 |
Workflow Clarity | The routing workflow (identify task → consult decision tree → read referenced file) is unambiguous and consistently encoded, and conflict-resolution ordering ('Use design FIRST... Then use swiftui') sequences cross-skill cases. Below 5 because multi-file chains are not explicitly sequenced and the four redundant renditions of the routing map must be kept mutually consistent. | 4 / 5 |
Progressive Disclosure | The body is structured exactly like anchor 5 — an overview with clearly signaled, one-level-deep references — but none of the referenced files (skills/hig.md, skills/liquid-glass.md, skills/sf-symbols-ref.md, etc.) exist in the bundle: there is no references/, skills/, or assets/ directory, so every pointer dangles in the evaluated context. Scored against the actual bundle structure, navigation cannot be verified and the intended content split is not realized, which pulls it to anchor 3. | 3 / 5 |
Total | 13 / 20 Passed |