Content
71%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.
The body is well-structured and actionable, with a clear fixed workflow, real feedback loops for empty/error cases, and excellent progressive disclosure via verified one-level-deep references. The main weakness is conciseness: the delegation and no-token constraints are restated multiple times and could be consolidated.
Suggestions
Consolidate the repeated 'Never directly execute CLI — delegate to ontology-core' and '无须 token / 不出现 --token' reminders into a single stated-once constraint block; reference it rather than restating verbatim in the body and every reference file.
Make the step-1-to-step-2 handoff checkpoint explicit (e.g., 'if step 1 yields no department/topic clues, fall back to the user's original terms or ask a clarifying question before calling the duty KN') instead of leaving it implied.
In the main body, show one minimal complete condition-json example inline so the command form is copy-paste-ready without requiring a jump to metadata-search.md.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly operational and free of concepts Claude already knows, but the delegation reminder ('Never 直接执行 CLI，委托 ontology-core') and the '无须 token / 不出现 --token' constraint are repeated verbatim several times across the body and each reference, which could be tightened. | 3 / 5 |
Actionability | Provides concrete command forms (`ontology --user-id <accountId> bkn object-type query <kn_id> <ot-id> '<condition-json>' [--limit n]`) and the references carry full, copy-paste-ready bash with complete condition-json bodies and example values; minor gaps come from the deliberate placeholder-only form in the main body. | 4 / 5 |
Workflow Clarity | A fixed, clearly sequenced 2-step-query + summarize pipeline with a progress checklist and feedback loops (empty result → rewrite `search`/switch KN/clarify; command error → report honestly, never fabricate); this is read-only retrieval so the destructive/batch validation cap does not apply, but an explicit 'verify step 1 returned clues before step 2' checkpoint is only implied. | 4 / 5 |
Progressive Disclosure | SKILL.md is a concise overview with a '必读 references（按步骤）' table signaling three one-level-deep reference files (metadata-search.md, duty-search.md, tool-examples.md), all verified present; content is appropriately split (overview / per-step command detail / end-to-end example) with easy navigation. | 5 / 5 |
Total | 16 / 20 Passed |