Content
92%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.
An excellent operational skill body: fully executable commands with real examples and failure receipts, a rigorously sequenced write workflow with validation and confirmation checkpoints, and clean one-level-deep reference structure. The only weakness is redundant restatement of the same safety rules across the IRON RULE, 注意事项, and Phase 0 sections.
Suggestions
Consolidate the duplicated safety rules: make 注意事项 reference the IRON RULE by number instead of restating rules 1–3 and 8, and trim the Phase 0 boundary list that repeats them again.
Move the 真实链路速记 worked example into a reference file (e.g., a new references/example-run.md) to further slim the always-loaded body.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with deployment-specific operational detail Claude could not know (e.g., the `product_entity\ LIMIT 50` sqlglot incident, `Error 1062` receipt shape, quote-stripping rules for `meta_table_name`) and explains no general concepts, fitting the 'efficient, minor trims possible' anchor. It falls short of lean/5 because the safety rules are restated three times — IRON RULE, 注意事项 items 1–3/8, and the Phase 0 sections repeat the same prohibitions. | 4 / 5 |
Actionability | Commands are fully executable and copy-paste ready with real IDs and expected outputs, e.g., `ontology ds import-csv 374c4c1b-8836-4e60-8099-572a1f0367c5 --table-name product_entity --file test_dedup.csv --batch-size 100` plus the exact failure receipt JSON and the expected HTTP endpoint, covering the common success and duplicate-key cases. | 5 / 5 |
Workflow Clarity | The 10-step ordered checklist (0a–10) with a copyable progress block has explicit validation checkpoints for a batch database write: three-stage target resolution, primary-key pre-check SQL, client-side diff-set removal, mandatory user confirmation of kn/object_type/dataview/datasource/table/rows, and receipt verification with per-row failure reporting and no blind retries — matching the anchor-5 feedback-loop pattern. | 5 / 5 |
Progressive Disclosure | The body is an overview that offloads per-step detail via a step→reference table; all five referenced files (parse-input.md, map-fields.md, resolve-target.md, csv-prepare.md, import-csv.md) exist and are one level deep, with their outbound links pointing to sibling skill contracts rather than nested detail. Navigation by step is easy and clearly signaled. | 5 / 5 |
Total | 19 / 20 Passed |