Content
75%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 operationally dense, highly actionable contract document with excellent command-level specificity, explicit validation and error-recovery rules, and a genuinely well-wired one-level-deep reference system. Its costs are token weight: duplicated constraint bullets, ~250 lines of root-level detail (much of it form-share contract that has a natural home in a reference), and unnamed bundle scripts.
Suggestions
Remove the verbatim duplicate bullets in 执行约束/记录稳定约束 (ID-read-back, hasMore/partial_success, --verbose/raw/pretty, timestamp-reuse) — each rule appears twice within ~20 lines and can be stated once.
Move the 表单分享用法回答契约 and 返回值评审专用查询 sections into a reference file (e.g., alongside references/aitable/aitable-form.md) and keep only the routing table plus the two-line answer template in SKILL.md; this cuts roughly a quarter of the root file while preserving the trigger rules.
Name and link the five scripts in scripts/ (aitable_import_via_task.py, upload_attachment.py, etc.) from the body — currently the runtime contract only says published scripts exist, leaving the bundle's most executable assets undiscoverable.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Almost all content is product-specific contract Claude could not know (routing tables, flag inventories, cursor-recovery rules), but there is real duplication: “复用 JSON 已返回字段,不以 --verbose/raw/pretty 重复请求” appears in both 执行约束 and 记录稳定约束 (lines 179/186), and “新增或更新只使用真实返回的 ID 回读…全量查询检查 hasMore…” is repeated verbatim within 记录稳定约束 (lines 196–197 vs 200–201), with the timestamp rule also stated twice (lines 178/185). This is the 'mostly efficient but could be tightened' anchor — noticeably better than the verbose padded anchor, but the literal duplicate bullets and the very dense 250-line root file keep it from 4. | 3 / 5 |
Actionability | Guidance is fully executable: complete commands with flags (`dws aitable +record-query --base-id <ID> --table-id <ID> [--record-ids <IDs>]…`), exact one-shot discovery commands (`dws schema --cli-path "aitable form share get" --compact --format json`), a copy-paste-ready two-line answer template with placeholder conventions, and concrete JSON shapes like `{"options":[{"name":"<选项>"}]}`. This matches the anchor for copy-paste-ready commands covering the common cases. | 5 / 5 |
Workflow Clarity | Multi-step flows are explicitly sequenced with validation checkpoints — the psql flow (`-l` discover → `-t` columns → `LIMIT 3` → `-c`), write-then-read-back-by-returned-ID, `hasMore`/`partial_success` checks, and a numbered 错误最短路径 with retry/stop rules, so it is above the 'most checkpoints, minor gaps' anchor. It falls short of 5 because the routing is a web of competing precedence rules (intent-routing table '优先于' schema navigation, the form-share gate '优先于' reference navigation, multiple 'not applicable to each other' carve-outs) that the reader must reconcile rather than one coherent sequence, and some checkpoints depend on cross-referenced references. | 4 / 5 |
Progressive Disclosure | Structure is good and real: all 23 cited reference paths resolve to existing one-level-deep files (references/*.md and references/aitable/*.md), and the 按需加载 table maps each trigger condition to exactly one reference with an explicit “不要预加载这些 Reference” guard — matching the good-structure anchor. It stops short of 5 because substantial deep procedural detail (the ~55-line 表单分享用法回答契约 and 返回值评审专用查询 sections, and the inline DWS runtime contract block) is inlined in the root SKILL.md instead of the existing references/aitable/aitable-form.md-style split, and the five scripts in scripts/ are mentioned only generically (“本 Skill 明确发布的脚本”) with no names or paths. | 4 / 5 |
Total | 16 / 20 Passed |