Content
81%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 highly actionable, well-structured diagnostic guide with executable code, prioritized root causes, explicit decision flows, and validation feedback loops. Its main weaknesses are redundancy — the animation loop and collider-reduction code each appear twice, plus speculative padding in Root Cause 2 — and the absence of any reference files for a ~290-line skill.
Suggestions
Remove the duplicated code in "Recommended Initialization Code": the animate loop repeats Root Cause 1's solution and the reduction loop repeats Option 1 — reference the earlier snippets instead.
Cut the speculative "Possible explanations" bullets and the VRChat tangent, and merge Root Cause 2B into Option 5, which already covers scene scaling.
Move the "Key API Reference" and "Joint Settings" tables plus the UniVRM evidence deep-dive into a reference file to slim the main SKILL.md.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly high-value domain content with no explanations of concepts Claude already knows, but noticeably loose in places: the animate loop appears in full twice (Root Cause 1 and "Recommended Initialization Code" — "delta = Math.min(delta, 0.05)"), the collider-reduction loop appears in both Option 1 and the init section, and there is speculative padding ("VRoid Studio (most common VRM source) may use consistent internal scaling", the VRChat tangent in Root Cause 2B). This is more than minor over-explanation (ruling out 4) but the padding is structural duplication rather than explained basics (ruling out 2). | 3 / 5 |
Actionability | Fully executable, copy-paste-ready code covers the common cases: THREE.Clock setup with delta clamping ("delta = Math.min(delta, 0.05); vrm.update(delta);"), collider reduction with original-radius preservation ("collider.shape.radius = collider._originalRadius * REDUCTION_FACTOR;"), a complete scaleVRMScene() function, diagnosis snippets ("if (c.shape?.radius) c.shape.radius = 0;"), concrete Unity fix steps, and API/joint settings tables. Only trivial gap: Option 3 uses `colliders` without restating its acquisition, which is shown in adjacent snippets — not enough to drop from anchor 5. | 5 / 5 |
Workflow Clarity | A clear prioritized diagnostic sequence with explicit validation checkpoints and feedback loops: root causes ordered by frequency ("Root Cause 1: Incorrect Delta Time (Most Common - 90%)"), a decision flow ("Only bangs horizontal → Head collider blocking them / All physics horizontal → Gravity direction wrong"), confirmation loops ("If hair now falls correctly, colliders were the issue"), and external validation ("Test with official VRM viewer - if hair is broken there, it's a model issue"). Not 4: validation is explicit for each major branch, not merely present with minor gaps. | 5 / 5 |
Progressive Disclosure | Good structure with well-organized sections and a clearly signaled one-level-deep external reference ("The file format, matching rules, known limitations... are in docs/frontend/vrm.md"), and most content (symptoms, diagnoses, fixes) appropriately lives in the main file. Not 5: no bundle reference files exist and at ~290 lines the "Key API Reference"/"Joint Settings" tables and the UniVRM evidence deep-dive ("#### Confirmed Facts", "#### Empirical Observation") are inline candidates for extraction; not 3: the structure is good and the inline content is mostly where it belongs. | 4 / 5 |
Total | 17 / 20 Passed |