Content
85%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.
Excellent operational content: copy-paste-ready commands, a disciplined shortcut-vs-raw-API decision path, and strong safety/validation workflows for destructive and batch mail operations, all properly externalized into 31 well-signaled reference files. The main cost is token weight from repeated safety rules and a partially redundant concept glossary in an already long body.
Suggestions
Consolidate the send-confirmation rule, currently stated in 安全规则 #5, the 数据真实性 section, the 命令选择 bullets, and the 默认话术 list — state it once authoritatively and cross-reference it elsewhere to cut duplicated tokens.
Merge the 草稿回链规则 (安全规则 #8) with the overlapping 默认话术 bullets at the end of 典型工作流, which repeat the same draft-link handling.
Trim the 核心概念 glossary to only Lark-specific semantics (message_id/thread_id, folder IDs, rule/template resources) and drop definitions of concepts Claude already knows, such as what an email, attachment, or reply thread is.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly operational and dense with non-obvious CLI specifics, but the send-confirmation rule is stated four times (安全规则 #5, the 数据真实性 intro, the 命令选择 bullets, and the 默认话术 list) and the draft-link rule twice, and the 核心概念 glossary partly explains concepts Claude already knows (what an email/attachment/thread is). It fits 'mostly efficient but could be tightened' rather than the minor-instances anchor 4. | 3 / 5 |
Actionability | Guidance is fully executable throughout: exact commands with flags ("lark-cli mail +send --to alice@example.com --subject '周报' --body '<p>...'"), copy-paste JSON for --params/--data, a table mapping schema fields to CLI flags, and a concrete 3-step raw-API procedure with worked GET/POST examples. | 5 / 5 |
Workflow Clarity | The 11-step 典型工作流 is clearly sequenced with explicit validation checkpoints: identity confirmation via profile, confirmation previews with affected counts for destructive/batch ops ("将删除 N 封邮件，确认？"), post-send send_status verification, optional +lint-html precheck, and mandatory -h discovery before any raw API call — including error-recovery guidance (fix and re-query). | 5 / 5 |
Progressive Disclosure | The SKILL.md is an overview plus quick-reference, with all 31 per-operation details split into one-level-deep references/ files that all exist and are clearly signaled in the shortcuts table and 'ref:' links; templates live in assets/templates/ (5 real HTML files). Navigation is easy and nothing nests more than one level. | 5 / 5 |
Total | 18 / 20 Passed |