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.
A strong, highly actionable performance skill: executable code for every lever, measurement-first framing, and valuable non-obvious details (counter reset semantics, Clock deprecation, shadow counter behavior). Its weaknesses are mild — some boilerplate in Quick Start, no explicit counter-to-lever diagnostic map, and no use of reference files for advanced material.
Suggestions
Trim the Quick Start boilerplate (renderer/scene/camera/light setup) down to the minimal loop that demonstrates reading renderer.info after render().
Add a short diagnostic table mapping high counter values to the matching section (e.g. info.render.calls high -> merge/instance/batch; info.memory high -> KTX2/dispose; frame time high with low CPU -> fill rate).
Move the WebGPU/TSL section and the KTX2 format detail into a references/ file, keeping a brief pointer plus the counter-name differences in the main body.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and assumes Claude's competence — no basic three.js or rendering explanations — with terse, high-value code comments (e.g. 'the counters reset at the start of every render() call'). The Quick Start block includes boilerplate renderer/scene/camera setup that could be trimmed, keeping it below the 'every token earns its place' anchor. | 4 / 5 |
Actionability | Every section provides copy-paste-ready executable code with specific APIs — renderer.info counters, mergeGeometries, InstancedMesh, BatchedMesh, LOD, KTX2Loader, compileAsync, shadow-map freezing, TSL compute — and the examples cover the common optimization cases end to end. | 5 / 5 |
Workflow Clarity | 'Measure First' establishes a clear diagnostic sequence, and verification checkpoints appear (checking info.memory after dispose, averaging Timer over 60 frames, separating CPU vs GPU time). A minor gap: there is no explicit counter-to-lever mapping that would guide choosing between sections when a specific number is high. | 4 / 5 |
Progressive Disclosure | The single file is well-sectioned and easily navigable, with clearly signaled one-level See Also links to sibling skills. However, all detail is inlined in a ~390-line file; advanced material such as the WebGPU/TSL compute example or the KTX2 format matrix could be split into reference files. | 4 / 5 |
Total | 17 / 20 Passed |