Content
62%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The skill is highly actionable with a clear, validated workflow and excellent executable SQL examples. Its weakness is conciseness — the sql_craft section is verbose and restates concepts Claude already knows — and progressive disclosure, since everything lives in one large inline file.
Suggestions
Move the bulk of the <sql_craft> heuristics into a reference file (e.g. SQL_CRAFT.md) and keep only the highest-priority rules inline, linking out for the rest.
Tighten each bullet to the rule plus a one-line rationale; cut restatements of basic SQL behavior (integer division, tie-breakers) that Claude already knows.
Consolidate the repeated 'Final completeness check' / 'Answer every requested output' guidance, which appears several times in slightly different forms, into a single checklist.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is ~260 lines of densely packed, repetitive guidance; many sql_craft bullets restate the same point at length and explain SQL mechanics Claude already knows, so a large share of tokens do not earn their place. | 1 / 3 |
Actionability | Provides concrete tool-call sequences, specific parameter shapes, and executable SQL examples (e.g. the parsed-volume CTE and the returned_orders pre-aggregation) that are copy-paste ready. | 3 / 3 |
Workflow Clarity | The numbered <workflow> lays out a clear discover→inspect→resolve→plan→query→validate→capture sequence with an explicit 'Always run the final completeness check before emitting' validation checkpoint and feedback loops for empty results. | 3 / 3 |
Progressive Disclosure | The body is a single monolithic file with no external references or bundle files, and the heavy sql_craft section (~150 lines) is inline where it could be split; sectioning is present but content that should be separate is not. | 2 / 3 |
Total | 9 / 12 Passed |