Content
47%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The skill is well-structured with a clear response workflow, checklist, and refinement loop, but it is severely over-long and repetitive for an instruction-only image-gen skill, with no progressive disclosure to offload the bulk. Actionability is mixed: a strong enumerated variation engine sits inside a lot of abstract feel-language.
Suggestions
Collapse the redundant sections (text readability appears in 28/29/30/35, phone mockup in 9/10/35/37, no-code in the directive/103-107/1385) into single canonical statements to cut length dramatically and raise conciseness.
Move the Style Variation Engine (Section 22), Category-Specific Bias (Section 33), and baseline config (Section 1) into separate reference files referenced one level deep from SKILL.md to improve progressive disclosure.
Replace abstract feel-directives ("feel premium", "be calm") with specific, checkable criteria so the guidance is more directly actionable.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~1460-line body is highly verbose and repetitive: the text-readability, phone-mockup, no-code, and anti-AI-tells directives each recur across multiple of the 38 sections, padded with feel-words ("premium", "calm", "believable") that restate mobile-design intuition Claude already has. | 1 / 3 |
Actionability | It offers concrete levers (the Style Variation Engine with "Choose exactly 4", numeric baseline config, explicit screen-count rules) but interleaves them with abstract direction ("feel premium", "be calm", "make it memorable") that is not directly executable. | 2 / 3 |
Workflow Clarity | Section 36 lays out a 28-step response sequence, Section 35 provides a 27-item quality checklist, and Section 34 defines a regenerate feedback loop — a clear sequence with checkpoints and an error-recovery loop matching the top anchor. | 3 / 3 |
Progressive Disclosure | The body is well-sectioned into 38 numbered headers, but it is a monolithic single file with no bundle/reference files; large reusable blocks (the variation engine, category biases, baseline config) that should be split out are all inlined. | 2 / 3 |
Total | 8 / 12 Passed |