Content
73%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 exceptionally actionable for an image-direction skill, with an explicit 16-step workflow, validation checklist, and error-recovery loops, and its section structure is easy to navigate. Its main weakness is redundancy: the core output rule and hero-bias guidance are each repeated verbatim in 3-4 places across 1012 lines, wasting context. No bundle references exist, so all guidance is inline, which is mostly appropriate but leaves some catalogs that could be split out.
Suggestions
State the one-image-per-section rule once (the HARD OUTPUT RULE preamble) and reference it from §5, §17, and §19 instead of repeating it verbatim four times.
Merge the duplicated hero-composition guidance — the "HERO COMPOSITION BIAS — READ FIRST" section and §4 "Hero Composition Bias" list nearly identical alternatives; keep one canonical list.
Fix the duplicated step numbering in §19 (two consecutive steps are numbered "5") and move the component catalog (§11) and slop lists (§8) into a reference file to reduce inline bulk.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The 1012-line body restates the same rules multiple times: the one-image-per-section rule appears in the "HARD OUTPUT RULE" preamble, again in §5 "IMAGE COUNT & PAGE SLICING" ("THIS IS THE PRIMARY OUTPUT RULE"), again in the §17 checklist items 19-20, and again in §19 step 16; the hero composition alternatives are listed nearly verbatim twice (the "HERO COMPOSITION BIAS" preamble and §4). Multiple padded passages ("Whitespace is a design tool. Use it deliberately. Do not let spacing become random.") add no instruction, matching anchor 2 ("several unnecessary explanations or padded sections") rather than 1 since almost all content is skill-specific rather than explaining things Claude already knows. | 2 / 5 |
Actionability | As an instruction-only skill, guidance is fully concrete and directive: exact numeric dials ("DESIGN_VARIANCE: 8"), enumerated one-pick options (theme paradigms, 12 background modes, 3 hero scales, 3 site packs with named sections), brief-to-direction mappings for specific user words ("minimalist", "editorial", "SaaS"), and three worked interpretation examples (§20) covering the common cases. The guidance is unambiguous enough to execute without inference, meeting the anchor-5 standard for instruction-only skills; not scored lower because no code is expected for an image-direction skill. | 5 / 5 |
Workflow Clarity | §19 lays out a numbered 16-step response workflow (infer site type → commit to section count aloud → choose scales/anchors per section → generate all labeled images, "Do not stop early"), backed by an explicit 21-item pre-output validation checklist (§17) and a feedback loop for error recovery ("If the count is wrong, regenerate the missing sections"; §18's "Reject the set if" criteria). Validation checkpoints and recovery loops are present for this batch multi-image operation, matching anchor 5; the only blemish is a duplicated step number ("5." appears twice in §19), which is a typo rather than a validation gap. | 5 / 5 |
Progressive Disclosure | No bundle files exist, and the body has a genuinely navigable structure: READ FIRST sections up top, 21 numbered and clearly titled sections, and a cross-referenced system (§1 references §2, §19 references §5/§17/§18). Most content is appropriately placed for a single-shot prompt skill, though the ~1000-line monolith with no external references means catalogs like the 12 signature components, slop lists, and default site packs could arguably live in reference files — a minor organization gap that keeps it at anchor 4 rather than 5's ideal split or 3's poorly-signaled inline bulk. | 4 / 5 |
Total | 16 / 20 Passed |