Content
88%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 highly actionable, dense operational skill: the Golden Route table plus command capsules give copy-paste-ready entry points, and validation/recovery discipline is exceptional for destructive and batch operations. The main costs are some duplicated rule text between the contract and shortcut sections, and a few reference paths that nest two levels deep.
Suggestions
Deduplicate the Help/Schema fallback rules, which appear almost verbatim in both the 最小 DWS 执行契约 and Shortcut 发现 sections — state them once and cross-reference.
Flatten navigation so all leaf references (e.g., 10/11-minutes-speaker-*.md, recipes/conventions.md) are reachable within one hop of SKILL.md's 按需加载 table or the index, rather than two hops via intent-guide.md/minutes.md.
Trim the recurring confirmation-gate statements in 安全边界 that restate rules already fixed in the runtime contract block.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense, operational, and explains nothing Claude already knows, but the Help/Schema fallback rules appear nearly verbatim in both the DWS runtime contract block and the Shortcut discovery block ("Schema 不可用才读已知 leaf Help 一次…`unknown command` 不查 Help…"), and confirmation rules recur across sections — minor duplication that could be trimmed. | 4 / 5 |
Actionability | The Golden Route table gives copy-paste-ready commands with concrete flags (e.g., `dws minutes +search --scope all --query "<关键词>" --page-all`), the search capsule is a runnable command block, and the timestamp script is invoked with an exact `python3 <本Skill目录>/scripts/format_timestamp.py --unit ms --timezone Asia/Shanghai <时间戳1>` line, covering the common cases. | 5 / 5 |
Workflow Clarity | Multi-step flows carry explicit validation checkpoints and recovery loops: pagination completeness must be proven by `data.complete=true`, writes must be verified after execution rather than by exit code, `unknown flag` triggers a one-shot Help correction, and async/partial failures preserve `taskUuid/taskId/sessionId/checkpoint` for targeted resumption; destructive and batch operations (撤权, 批量授权) require confirmation and per-member ledgers with failure evidence, so no validation cap applies. | 5 / 5 |
Progressive Disclosure | SKILL.md is an overview with a trigger-to-reference table (复杂流程/局部意图/原子命令) and a hub index (references/minutes.md) with back-links organizing the bundle, but some material sits two hops deep (SKILL.md → intent-guide.md → 10/11-minutes-speaker-*.md; → minutes.md → recipes/conventions.md), short of the fully one-level-deep ideal. | 4 / 5 |
Total | 18 / 20 Passed |