Content
75%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 body with executable pinned-version code, explicit validation and error-recovery guidance, and a properly signaled one-level reference. Its main weakness is token efficiency: an introductory explanation of what UMAP is and a parameter guide that duplicates the reference file add avoidable tokens.
Suggestions
Cut the overview sentence explaining what UMAP is ('UMAP (Uniform Manifold Approximation and Projection) is a dimensionality reduction technique...') — Claude already knows this — and lead with when to apply the skill.
Compress the Parameter Tuning Guide's 'Purpose/How it works/Effects by value' prose to a compact value-range table, since the full parameter contracts already live in references/api_reference.md; drop the duplicated metric enumeration from SKILL.md.
Add a validation checkpoint (e.g. verify finite coordinates and neighborhood retention) to the Typical Workflow so the primary getting-started path matches the checkpoint quality of the clustering workflow.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The opening paragraph explains what UMAP is — a concept Claude already knows — and the Parameter Tuning Guide's 'Purpose/How it works/Effects by value' prose plus the metric enumeration substantially duplicate the same parameter ranges and metric lists in references/api_reference.md. Most content is dense and library-specific, so it is not 2, but the redundant tutorial-style explanation and duplication could be trimmed, which fits anchor 3 rather than 4. | 3 / 5 |
Actionability | Guidance is fully executable throughout: pinned install commands ('uv pip install umap-learn==0.5.12', 'hdbscan==0.8.44'), copy-paste-ready fit/transform/clustering/pipeline/persistence code, and a declared placeholder convention ('data is a finite samples-by-features array and labels must follow the same row order') make the examples covering the common cases directly runnable. Minor undeclared names like input_dim follow the same convention and do not constitute gaps. | 5 / 5 |
Workflow Clarity | Multi-step processes are clearly sequenced with an explicit '4. Evaluate' step in the clustering workflow, a 'Validation:' paragraph, held-out evaluation in the ML pipeline section, and a Common Issues → Solutions recovery table. It is not 5 because the main Typical Workflow has no validation checkpoint and there are no checklists; it is well above 3 because checkpoints and error-recovery guidance are explicitly present in the substantive workflows. | 4 / 5 |
Progressive Disclosure | Structure is good: one clearly signaled, one-level-deep reference (references/api_reference.md) whose cited sections 'AlignedUMAP Class' and 'Usage Examples' exist, with a Resources section directing when to load it. It is not 5 because the parameter and metric detail inlined in SKILL.md overlaps content that lives in the reference file; it is not 3 because references are well signaled, real, and shallow. | 4 / 5 |
Total | 16 / 20 Passed |