Conciseness

The rules are maximally dense with exact formulas and constants ("α(s) = a·(sinθ + 0.16 sin(2θ+0.8))", "sheen 0.5–1.0", "normalBias 0.012") and assume Claude's competence with no padding about what Three.js or cloth is. Not 5: the KIBORI provenance paragraph ("Extracted from KIBORI 木彫, a dark WebGL workshop...") and the closing self-assessment ("Score against the rubric ... Overall 6.5") are meta-content that does not instruct the reader and could be trimmed.

Actionability

Mostly executable: exact GLSL/math expressions ("warp height blends ... by 0.5·smoothstep(0.05, 0.5, |t|)", "twill 2/1 = (A+B)%3 < 2"), a per-fabric constants table, and copy-ready parameter values. Not 5 because there is no complete copy-paste code block — the pieces must be assembled — and "demo.html is the reference" points to a file that is not present in the bundle, leaving the working example missing.

Workflow Clarity

The 13 numbered rules each pair an instruction with the failure it prevents, and the "Gotchas, symptom first" section gives explicit symptom → cause → fix recovery loops (e.g. "Weave crawls or shimmers → uv was rebuilt from arc length; keep it as the rest metres"). Not 5: there is no explicit build sequence (material → bake → geometry → light) tying the rules into an ordered procedure, and visual validation checkpoints are implicit in the gotchas rather than stated steps.

Progressive Disclosure

Sections are well organized (rules, constants, gotchas, lifecycle, cost), but the only bundle reference — "demo.html is the reference" — points to a file that does not exist (no references/, scripts/ or assets/ directories ship with the skill), so navigation dead-ends. This matches the middle anchor: structure is present, but the reference situation is broken and the full implementation the body leans on is missing. Not 4 because a claimed one-level reference that is absent is worse than minor organization gaps.

Specificity

The description lists many concrete capabilities: "thread-level weave normal map, sheen and anisotropic highlights, folds that never stretch the texture, hems with thickness, stitches, seams, frayed edges, fold occlusion, dye and wear variation, and light passing through thin fabric" — comprehensive and fully concrete, matching the anchor for multiple specific concrete actions. Not 4 because coverage of what the skill builds is essentially complete with no generic filler.

Completeness

Both questions are answered explicitly: the what is "Build convincing woven cloth in Three.js" with a concrete feature list, and the when is an explicit "Use for linen, silk, wool, canvas, curtains, banners, cloaks, tablecloths, tapestries or any 3D drape that currently looks like..." clause with concrete trigger phrases. This matches the top anchor exactly; voice is third person ("Build"), so no penalty applies.

Trigger Term Quality

Natural object and material nouns users would say are well covered: "linen, silk, wool, canvas, curtains, banners, cloaks, tablecloths, tapestries" plus symptom phrases like "looks like a plastic sheet, a stretched texture, or a flat coloured plane". Not 5 because common synonyms such as "fabric", "textile", "flag" or "garment" are missing, leaving a few natural trigger phrasings uncovered.

Distinctiveness Conflict Risk

The niche is clear — woven and sewn fabric rendering in Three.js — with distinctive symptom-based triggers ("plastic sheet", "stretched texture", "flat coloured plane") that would not fire for wood, metal or paper material skills. Not 4: no meaningful overlap risk remains; the material and object list is unambiguously cloth-specific.