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 dense but high-value persona skill: almost every section supplies Zhang-Xuefeng-specific data, an agentic research-before-answering protocol with explicit checkpoints and fallbacks, and concrete style rules with worked examples. The only real costs are inline duplication of research-source material that belongs in references/ and a verbose version-check section.
Suggestions
Conciseness/progressive_disclosure: move the 附录：调研来源 book/media lists and repeated key quotes into `references/research/` and keep a one-line pointer — the mental-model sections already cite the essential evidence inline.
Conciseness: compress the 版本检查 section, which spells out full user-facing message templates; a short table of conditions plus one message template would preserve the behavior at a fraction of the tokens.
Progressive disclosure: either reference `assets/hero.gif` from the body (e.g. in an author/about section) or remove it, since it is currently dead bundle weight.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Nearly all of the body is persona-specific knowledge Claude cannot know (e.g. 「社会就是一个大筛子，用学历筛孩子」quotes, the fallback tree, 表达DNA rules), so tokens largely earn their place. Minor padding exists: the 附录：调研来源 source lists duplicate evidence already cited in the 模型 sections, and the 版本检查 section spells out full user-facing message templates at length. | 4 / 5 |
Actionability | Fully executable guidance for an instruction-only skill: exact workflow directives (「⚠️ 必须使用工具（WebSearch等）获取真实信息，不可跳过」), a concrete git command (`git -C <本目录> ls-remote --tags ...`), a 9-row fallback table with first-choice and backup actions, 8 anti-patterns with corrections, and worked baseline-vs-round1 example dialogues covering common cases. | 5 / 5 |
Workflow Clarity | Clear Step 1→2→3 sequence (问题分类 → 张雪峰式研究 → 回答) with an explicit validation checkpoint (「🔴 CHECKPOINT · 开口前三问」) and a feedback loop (「任意一项答『否』→ 回到对应 Step，别硬出」), plus a failure-mode table mapping trigger signals to repair paths — exactly the validation/recovery anchor. | 5 / 5 |
Progressive Disclosure | References are real files, one level deep, and clearly signaled with conditional-read rules (「调研过程详见 `references/research/` 目录」; read `references/about-author.md` only when asked about the author). Not a 5: the 附录：调研来源 quotes and source lists duplicate reference content inline, and `assets/hero.gif` exists but is never referenced from the body. | 4 / 5 |
Total | 18 / 20 Passed |