Content
86%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 strong, operator-grade CLI skill body: dense with genuinely non-obvious deployment knowledge, executable commands, and a clean progressive-disclosure structure routing detail to real, well-organized reference files. The main improvements are deduplicating the repeated 'unavailable command groups' and --user-id warnings, and surfacing at least one inline build workflow with an explicit validation checkpoint.
Suggestions
Consolidate the 'unavailable command groups' (auth/config/agent/skill/toolbox/dataflow) warning, which currently appears three times (命令组总览 note, Metric-vs-logic_properties routing section, and 注意事项), into one canonical location.
State the --user-id <accountId> requirement once (in 调用约定) and reference it from other sections instead of restating the rule and its failure modes in multiple places.
Inline a compact version of the build-KN workflow's validation step (e.g. 'after bkn push/build, verify with object-type query limit 1 before reporting success') so the destructive/batch path has an explicit checkpoint in SKILL.md itself.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Nearly every token is non-obvious operational knowledge (jq top-level structure table, gotcha matrix, Metric vs logic_properties routing), but the 'unavailable command groups' fact is repeated three times (command-group overview note, routing section, and 注意事项) and the --user-id requirement is restated across sections — minor over-explanation that could be trimmed, matching the 4 anchor rather than 5. | 4 / 5 |
Actionability | Guidance is fully executable: copy-paste-ready commands (e.g. the jq projection pipeline for object-type list), correct-vs-wrong JSON condition shapes, explicit legal operator values, and a self-documenting escape hatch (ontology <group> <subcommand> --help) covering the common cases — matching the 5 anchor. | 5 / 5 |
Workflow Clarity | The query workflow has a clear sequence with a built-in checkpoint ('先裸跑一次 | jq keys 确认顶层键名再决定入口') and error handling (401 → report directly), but the multi-step build-from-DB workflow with its validation steps lives entirely in references/build-kn-from-db.md rather than the body, and truncation-risk batch listings get mitigation advice without an explicit verify-after step — clear sequence with minor validation gaps, the 4 anchor. | 4 / 5 |
Progressive Disclosure | The body is a well-signaled overview: a command-group table routes each domain to a one-level-deep references/*.md (all seven files exist, and both anchor links in the body resolve to real headers in references/bkn.md), with explicit on-demand reading guidance ('按需阅读') — matching the 5 anchor's clear overview with one-level-deep references. | 5 / 5 |
Total | 18 / 20 Passed |