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-referenced skill with executable examples, sensible validation checkpoints, and a genuine progressive-disclosure structure. Its main weakness is redundancy: format rules and the `--empty` flag are each explained multiple times, and reference listings are repeated, inflating token cost without adding information.
Suggestions
Consolidate the format guidance into one place: the 'How to choose the `format`' table plus the two 'Critical `sql`' notes and the trailing 'Notes:' list repeat the same dict/csv/sql rules three times — keep the table and the single critical ephemeral-model warning, cut the rest.
Merge '### The `--empty` flag' section into workflow step 4 ('Ensure upstream models exist before running'), which already covers the same flag with the destructive-data warning; delete the duplicate explanation and command example.
Remove the duplicate warehouse/special-case reference link lists in 'Platform/adapter-specific caveats' and 'Platform/adapter-specific data types' (or reduce them to only the caveat files, deferring data-type pointers to Additional Resources), and drop filler lines like 'Self explanatory -- the title says it all!'.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly dbt-specific knowledge Claude does not know, but contains real redundancy: the ephemeral/requires-sql-format rule is stated three times (Handy to know, the Critical `sql` note, and Notes:), the `--empty` flag is explained both in workflow step 4 and again in a dedicated later section, the format-selection table is restated in Notes:, and filler like "Self explanatory -- the title says it all!" adds nothing. This matches 'mostly efficient but could be tightened'; not 4 given multiple duplicated sections. | 3 / 5 |
Actionability | Copy-paste-ready YAML for both a minimal and a realistic test, executable commands ("dbt build --select hello_world", "dbt test --select \"hello_world,test_type:unit\"", "dbt run --select +my_model --empty"), a format-choosing table, and a Common Mistakes table with fixes fully cover the common cases. Matches the anchor-5 'fully executable, common cases covered' example; nothing below 5 is warranted. | 5 / 5 |
Workflow Clarity | A clear 4-step sequence (choose model, mock inputs, mock output, ensure upstream exists) includes a validation checkpoint for upstream existence, an explicit destructive-operation warning on `--empty`, and a failure-recovery section interpreting the actual-vs-expected diff. Not 5 because the workflow is fragmented: execution guidance and the `--empty` material are split across distant, partially duplicated sections rather than one coherent flow with explicit feedback loops. | 4 / 5 |
Progressive Disclosure | Thirteen real, one-level-deep reference files (all verified present in references/), clearly signaled with descriptions in the Additional Resources list and inline pointers in context. Not 5 because the warehouse and special-case reference listings are duplicated in two later sections and the ~390-line body inlines some material (e.g., the full email example) that could live in references/examples.md. | 4 / 5 |
Total | 16 / 20 Passed |