Content
46%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 rich, well-organized, and highly actionable seaborn reference with strong code examples and good plot-selection guidance, but it fails progressive disclosure: it is a ~660-line monolith whose promised reference files are missing from the bundle, and much of the catalog restates API knowledge Claude already has. It should be trimmed to an overview plus decision guidance, with the reference material actually moved into the referenced files.
Suggestions
Create the promised references/ files (function_reference.md, objects_interface.md, examples.md) and move the per-function parameter catalogs, palette listings, and theming listings out of SKILL.md so they stop being dangling references.
Cut conceptual padding Claude already knows — the Design Philosophy section and the palette/style/context name inventories — and keep the genuinely high-value material: figure-level vs axes-level distinction, long/wide-form data guidance, the plot-selection table, and Troubleshooting.
Add an explicit end-to-end workflow (prepare long-form data -> select plot type via the data-shape table -> map semantics -> theme -> save) so the sequence does not have to be inferred from catalog sections.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | At ~660 lines, the body catalogs seaborn's well-documented API inline — e.g. "Seaborn follows these core principles: 1. Dataset-oriented... 2. Semantic mapping...", full parameter lists ("stat - Normalization: 'count', 'frequency', 'probability', 'density'"), and palette/style name inventories ("rocket", "mako", "darkgrid", "paper"). This matches the anchor 'noticeably verbose; several unnecessary explanations or padded sections': most of the catalog restates knowledge Claude already has and duplicates material the Resources section says lives in reference files. Not score 3 because the padding is whole sections rather than a few trimmable spots; not score 1 because there is no padded narrative prose and the per-section formatting is tight. | 2 / 5 |
Actionability | Concrete, mostly executable code throughout: Quick Start ("df = sns.load_dataset('tips'); sns.scatterplot(data=df, x='total_bill', y='tip', hue='day')"), per-category examples, a plot-selection decision table ("Continuous x, continuous y: scatterplot, lineplot, kdeplot, regplot"), and Troubleshooting fixes ("sns.kdeplot(data=df, x='x', bw_adjust=0.5)"). Minor gaps keep it below the fully copy-paste-ready anchor 5: some snippets depend on undefined variables (timeseries, data in clustermap), and one fix uses a private API ("g._legend.set_bbox_to_anchor"). | 4 / 5 |
Workflow Clarity | No explicit sequence is written out; the natural pipeline (prepare long-form data -> choose plot type -> map semantics -> theme -> save) must be inferred across sections. The numbered Best Practices, the 'Choose the Right Plot Type' mapping, the Common Patterns end-to-end recipes, and the Troubleshooting error-recovery section provide partial sequencing, but checkpoints are absent and steps remain implicit — matching anchor 3 ('sequence present but checkpoints missing or implicit'). Not 4 because no clearly sequenced workflow is actually written; the destructive/batch validation cap does not apply since plotting is non-destructive. | 3 / 5 |
Progressive Disclosure | The Resources section promises "function_reference.md", "objects_interface.md", and "examples.md" under references/, but no references/ directory exists in the bundle — these are dangling pointers. Meanwhile the body inlines roughly 400 lines of function, palette, and theming reference that clearly belongs in those files, matching anchor 2 ('content that clearly belongs in separate files is inlined'). Not 3 because the references are not merely unclearly signaled — they do not exist; not 1 because section headers provide real structure and there is no nested-reference chain. | 2 / 5 |
Total | 11 / 20 Passed |