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 thorough, well-structured instruction skill that gives specific actionable guidance across data shapes, severities, and reporting. Its main weaknesses are mild redundancy between the workflow and standards sections, no executable code templates, and a single-file layout longer than ideal for progressive disclosure.
Suggestions
De-duplicate the check guidance between Workflow steps 4-5 and the Standards 'Core Checks'/'Specific Check Guidance' sections so each check type appears in one canonical place.
Add one or two short executable SQL/Python snippets for the most common checks (e.g. duplicate primary-key rate, null rate by partition) to lift actionability from concrete guidance to copy-paste ready.
Move the detailed 'Specific Check Guidance' catalog into a references/ file (e.g. CHECKS.md) and link to it from SKILL.md, keeping the body as a concise overview to improve progressive disclosure.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is a dense, high-signal catalog of checks, severities, and output structure with no basic-concept padding, but the check catalogs overlap between workflow steps 4-5 and the Standards section and the 'Compare rates, not just counts' guidance is restated, leaving minor trim opportunities. | 4 / 5 |
Actionability | Guidance is concrete and specific — exact checks per data shape, a severity rubric, a report structure, and named tools ($jupyter-notebooks, structured_data, operations_logs) — but it is instruction-only with no copy-paste SQL/code templates, so it stops at 4 rather than 5. | 4 / 5 |
Workflow Clarity | An unambiguous 8-step sequence runs from clarifying context through recommending fixes, with soft checkpoints ('Confirm grain before interpreting anomalies', label assumptions); it lacks an explicit validate-and-retry feedback loop, but the work is read-only analysis so the destructive/batch cap does not apply. | 4 / 5 |
Progressive Disclosure | Sections are well-organized with clear cross-skill and tool pointers, but at over 160 lines the detailed Core/Specific check catalogs are inlined with no bundle files to offload them, which is a minor organization gap rather than ideal one-level-deep splitting. | 4 / 5 |
Total | 16 / 20 Passed |