Content
82%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 a strong, lean reference for a domain-specific environment: every operation is executable via concrete ask_env examples, domain concepts (AOI/POI IDs, modes, categories) are documented compactly, and risky behavior is flagged. The only meaningful gaps are the absence of explicit validation/recovery loops and the lack of any progressive-disclosure split despite reference-worthy detail tables.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is token-efficient: table-formatted 移动模式 and POI 类别 data, terse per-tool parameter lists, and no padding or explanation of concepts Claude already knows. The ~20-line category table and repeated ask_env boilerplate could be trimmed slightly, keeping it at anchor 4 rather than 5. | 4 / 5 |
Actionability | It gives the exact call syntax (ask_env(instruction=..., ctx={"id": ...})), complete parameter lists for every tool, and copy-paste-ready example instructions with expected outputs covering the common cases (observe, commute, search-and-travel, return home) — matching the 'fully executable' anchor. | 5 / 5 |
Workflow Clarity | Sequences are clearly demonstrated through worked examples (observe → recommend → move) and the 约束 section explicitly flags the risky operation of interrupting an in-progress move_to. Checkpoints are implicit (idle/moving status) rather than explicit validate-and-recover loops, so it sits at anchor 4 rather than 5. | 4 / 5 |
Progressive Disclosure | As a single-file skill with no bundle files, it is well-structured: clear section headers, consistent tables, and a closing constraints section make navigation easy. At ~130 lines the under-50-line exception does not apply, and the category/tool reference detail could be split into a separate file, matching anchor 4 rather than 5. | 4 / 5 |
Total | 17 / 20 Passed |