Content
100%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 exemplary operational skill body: lean and dense, fully executable commands with defined variables, a validated multi-step workflow with refusal-based checkpoints and recovery paths, and a well-structured one-level-deep reference bundle. No dimension shows more than a trivial gap.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense operational prose with no padding and no explanations of concepts Claude already knows; even the JS-portal rendering note is domain-specific knowledge (job.10086.cn etc. return empty shells) that Claude would not have. Every section carries rules or commands that earn their tokens. | 5 / 5 |
Actionability | Guidance is copy-paste executable: concrete commands with defined variables ("python3 \"$S/radar.py\" --root \"$ROOT\" doctor", "radar.py --root \"$ROOT\" --run <run> init --plan <that file>", "collect-ats --config source.json", "ingest --input observations.json", "radar.py watch --add <entry.json>") plus named plan fields (scope, instances, max_records, priority) and concrete file paths. | 5 / 5 |
Workflow Clarity | The procedure is a clearly numbered doctor → init → collect → screen → export → closeout sequence with explicit validation checkpoints and feedback loops: closeout refuses runs missing quota_note, watch_missing entries, current rules, or a current workbook; failures are reported as "not complete" with reasons; amend/resume paths handle mid-run changes and lease contention. Batch operations are well covered by validation, so no cap applies. | 5 / 5 |
Progressive Disclosure | Verified against the actual bundle: all six references cited in the body (hosts.md, record-schema.md, feishu-source.md, collection.md, portals.md, jev.md) exist in references/, are one level deep with no nested references, and each link is signaled with when to read it; detailed schemas and scripts (radar.py, initial_filter.py, workbook.py) are properly deferred out of the overview. | 5 / 5 |
Total | 20 / 20 Passed |