Content
71%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 dense, expert-level build guide with rare concreteness — measured constants, exact shader expressions, and symptom-first gotchas — that mostly respects the reader's intelligence. Its main weaknesses are bundle hygiene: the demo and source files it tells the reader to read are absent, there is no complete code in the skill itself, and a couple of meta sections spend tokens on non-guidance content.
Suggestions
Ship demo.html (and kibori.html or a distilled reference) in the bundle, or inline the essential onBeforeCompile patch code — the body's key instruction is 'Read demo.html for a working build' and that file is not there.
Move the self-score table and provenance section into a reference file (or trim them) — they are self-assessment, not build guidance, and cost roughly 13 lines of context.
Add one explicit validation checkpoint per pipeline stage (e.g. 'at overview, confirm the tear reads torn, not wavy, before enabling macro') so the error-recovery loops in Gotchas attach to the sequence rather than living apart from it.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and assumes competence — no explanation of what Three.js, height maps, or albedo are, and every rule is paired with the failure it prevents ("Rules and the failure each prevents"). It is not a 5 because the "Self-score against the anchored rubric" table and "Provenance" section (~13 lines combined) are self-assessment/meta content rather than build guidance, and the "Playback and lifecycle" section overlaps knowledge Claude already has (document.hidden, DPR caps, ResizeObserver), so a few tokens could be trimmed. | 4 / 5 |
Actionability | Highly concrete throughout: exact formulas ("tr = 0.74 + (0.5-form)*0.85 + (0.5-lowfbm)*0.45 - fibre*0.22 - fleck*0.45"), working GLSL expressions (the smoothstep ink-bleed and ellipse-fleck snippets), a full constants table per stock, and specific gotcha fixes. It stops short of anchor 5 because there is no complete executable code — the reader is pointed to "demo.html for a working build", and that file (along with kibori.html) is not present in the bundle, leaving the copy-paste-ready anchor missing. | 4 / 5 |
Workflow Clarity | The 7-step pipeline is clearly sequenced (bake, warp, mask, shape, shade, transmit, macro) with exact parameters at each step, and the symptom-first "Gotchas" section provides genuine error-recovery feedback loops (e.g. fringe fibres vanish → alphaTest explanation → alphaToCoverage fix). Not a 5 because there are no explicit validation checkpoints between stages (e.g. no 'verify the tear reads torn at overview before proceeding to shading'), so checkpoints remain implicit rather than stated. | 4 / 5 |
Progressive Disclosure | Sections are well-organized and clearly headed, but the body references "demo.html" and "kibori.html" as load-bearing materials ('Read demo.html for a working build', 'ported unchanged from the Kibori workshop page') and neither file exists in the bundle — no references/, scripts/, assets/ directories are present. Per the guideline to score against the actual bundle structure, these are dangling references to files that cannot be read; the constants table and gotchas are also candidates for a separate reference file given the >50-line length. | 3 / 5 |
Total | 15 / 20 Passed |