Content
96%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 an exemplary operational skill: fully executable commands, pervasive validation gates and feedback loops around destructive messaging operations, and dense constraint prose with zero padding. The only structural weakness is that a long protocol-construction paragraph is inlined in SKILL.md instead of being offloaded to the already-referenced protocol document.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | There is no padding and no explanation of concepts Claude already knows; every sentence is an operational constraint ("列表 cursor 是页码，历史 cursor 是上一页真实 store_id", "提交结果不明先查历史，不直接重发", "群聊拒绝发送、已读、撤回和删除"). Every token encodes skill-specific protocol or safety knowledge, matching the lean/efficient anchor. | 5 / 5 |
Actionability | Copy-paste-ready commands with real flags cover all common cases: "xhs-im list --limit 30 --cursor 0", "xhs-im history --conversation group:群ID", the guard invocations "python3 <guard路径> check --conversation 用户ID --history-file … --ledger …", and the raw API escape hatch "xhs-im call send_short_link_message --kwargs '{…}' --confirm". This matches the fully-executable anchor. | 5 / 5 |
Workflow Clarity | Destructive write operations are gated by explicit validation and feedback loops: preflight "xhs-hunter check", preview-by-default with "--confirm", safe-stop rules ("缺会话…时安全停止，不用默认零伪造回执"), and the guard lifecycle record-unknown → send → submitted → verify history → confirmed with stop conditions. The destructive-operation cap of 3 does not apply because validation is pervasive, matching the anchor with explicit validation steps and error-recovery loops. | 5 / 5 |
Progressive Disclosure | Sections are well-organized (读取 / 写操作 / 自动回复前置判断) and external pointers are one level deep and clearly signaled ("_shared/xhs_im_reply_guard.py" with its relative path, "docs/xhs-im-read-protocol-2026-10-24.md"). However, the dense read-protocol construction paragraph ("已读请求按网页 IM 客户端的实际构造逻辑生成…max_store_id…info.is_friend / info.is_mute…") is inlined rather than summarized with details left to the referenced protocol doc, and the body exceeds the ~50-line simple-skill threshold — minor organization gaps rather than a fully appropriate split. | 4 / 5 |
Total | 19 / 20 Passed |