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 well-organized, appropriately terse instruction skill: the mandatory render-inspect-revise workflow with its gated self-review checklist is exemplary, and the bundled mplstyle asset is genuinely used rather than duplicated. The main gaps are minor: no copy-paste-ready code snippet, some duplicated export guidance, and a trailing section of external policy pointers that are not part of the bundle.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is a lean, well-sectioned list of rules with almost no explanation of concepts Claude already knows; minor trimmable spots remain, such as the motivational prose in 'Core principle' ("The goal is not to make a plot 'fancy'."), the vague "Do not rely entirely on defaults", and the duplicated export-format guidance across 'Surface classes', 'Style contract', and 'Export discipline'. This fits level 4 ('efficient; minor instances that could be trimmed') rather than level 5, and is clearly above level 3 since padding is the exception, not the rule. | 4 / 5 |
Actionability | Guidance is concrete and specific for an instruction-only skill: a real bundled style path (assets/deepscientist-academic.mplstyle, verified to exist), exact figure sizes in inches, per-class export formats, an explicit inspection workflow, and a verbatim caption sentence to include. It falls short of level 5 because the only code-shaped guidance is a placeholder pattern (`plt.style.use(<path-to-assets/deepscientist-academic.mplstyle>)`) rather than a copy-paste-ready snippet, and some rules (e.g. 'legend minimal') leave the how to Claude. It is above level 3 because the guidance is direct and executable in substance, not pseudocode. | 4 / 5 |
Workflow Clarity | The mandatory workflow is a clearly numbered 5-step sequence with an explicit validation step ("open the rendered figure yourself... inspect the actual result, not just the plotting code"), a feedback loop ("revise the figure if readability or composition is weak" then "re-export"), and a 10-item self-review checklist gated by "If any answer is negative, revise before calling the figure complete." This matches level 5 exactly: clear sequence, explicit validation, feedback loop, and a checklist; the operation is not destructive/batch so no capping applies. | 5 / 5 |
Progressive Disclosure | The body is well-sectioned and the single bundle asset is real, appropriately referenced by path, and correctly kept out of the main file (the .mplstyle carries the concrete style values rather than duplicating them inline). It stays below level 5 because the skill is ~180 lines with no references/ split — the chart-selection heuristics, color rules, and size tables are candidates for a reference file — and the 'References to internal policy' section points to six paths that are not part of this bundle and are not clearly signaled as to whether they must be read. | 4 / 5 |
Total | 17 / 20 Passed |