Content
82%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.
An expert-level, highly dense procedural-geometry skill: no filler, hard-won failure modes surfaced as bolded callouts, and a closing verification checklist. The main gaps are two stubbed code bodies in the core roof and UV functions and the absence of an explicit fix-and-re-check loop around the final checks.
Suggestions
Complete the stubbed return in roofPoint() (the lerp from inner edge to outer edge scaled by flare) and the UV rewrite loop body, since these are the two functions the whole roof character and scaffolding realism depend on.
Turn the 'What to check before you call it done' list into an explicit fix-and-recheck loop (e.g. 'if walls are missing, flip winding on the prism side faces and re-orbit') so verification drives correction, not just diagnosis.
Consider moving the builder API signature table and the per-style caps/stages tables into a reference file so SKILL.md stays an overview of technique.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Lean and efficient throughout: dense imperative prose with no explanation of concepts Claude already knows, and every token carries non-obvious domain knowledge ('Merge by material, not by part. Six buildings drawn in 11–18 draw calls is the difference between 120fps and 30'). | 5 / 5 |
Actionability | Mostly executable guidance — the clipping-plane snippet, cap geometry table and stages table are concrete with real numbers — but two core code blocks have stub bodies: 'return [ /* lerp inner edge → outer edge × flare */ ]' and '/* scale by real size, offset randomly */', leaving key details to be filled in. | 4 / 5 |
Workflow Clarity | Sections follow a clear build order (vocabulary → roofs → plans → detailing → animation → scaffolding → timeline) and end with a 'What to check before you call it done' checklist that pairs symptoms with causes ('Missing walls are inverted winding'). Minor gap: no explicit fix-and-re-check feedback loop around the verification steps. | 4 / 5 |
Progressive Disclosure | Good structure for a single-file skill: well-headed sections, clearly signaled sibling-skill pointers ('Stand it in threejs-landscape and weather it with threejs-weather'), no nested references. The builder API table and per-style level/stage tables are candidates for separate reference files but the document stays navigable. | 4 / 5 |
Total | 17 / 20 Passed |