Content
53%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 SKILL.md body is punchy and dense, with a well-specified answer format and unusually good failure-mode documentation, but it is undermined by its bundle: both files it delegates its core workflow and design patterns to are missing, leaving the actual procedure only sketched in examples. Add an internal inconsistency (three-part vs. four-part answer) and the skill as shipped is more manifesto than executable instruction.
Suggestions
Ship Workflows/Extract.md and References/QuestionDesign.md in the bundle (or inline their essential steps and question-design patterns into SKILL.md) — the workflow section currently points at a file that does not exist.
Fix the internal inconsistency: Example 1 says "写 A 三段" while 铁律 2 mandates a strict four-part structure (结论/形式化/论证步/边界) — align the examples with the rule.
Promote the rewrite-after-scan loop from Gotchas into an explicit numbered workflow step with a validation checkpoint (generate → scan Q list → rewrite definition-answerable Qs → verify each A has all four parts).
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and assumes Claude's competence — no basic-concept explanations, and every Gotcha is a genuine failure mode ("AI 默认会写「什么是 X」型问题"). Minor trimming is possible: the nail/chisel metaphor repeats three times ("每个 A 砸下来一枚钉子", "Q 是凿子,A 是钉子", rule 3), and "你不是"/"你是" restate one framing from two sides — that places it just below anchor 5. | 4 / 5 |
Actionability | There is concrete guidance (the strict four-part A structure "*结论* + *形式化* + *论证步* + *边界*" with worked formalization examples like `通才 = 协调,专才 = 干活`, a copy-paste curl notification command, and a concrete denote filename pattern), but the core extraction procedure is delegated to "`Workflows/Extract.md`", which does not exist in the bundle — a key missing detail. This sits between anchor 3 (incomplete, missing key details) and anchor 4 (minor gaps only); the absent workflow file is a major gap, so 3. | 3 / 5 |
Workflow Clarity | A rough sequence appears only in the Examples ("WebFetch 获取 → 找观点骨架 → 设计 Q 链 → 写 A 三段 → markdown 输出") and a revision loop exists in Gotchas ("生成后扫一遍,凡是 Q 能用一句定义打发的,重写"), but the promised step list ("按 `Workflows/Extract.md` 的步骤执行") points to a file that is not in the bundle, and Example 1's "写 A 三段" contradicts rule 2's four-part mandate. Sequence present with real gaps and an internal inconsistency = anchor 3, not the mostly-checkpointed flow of anchor 4. | 3 / 5 |
Progressive Disclosure | The bundle contains only SKILL.md — neither "`Workflows/Extract.md`" nor "`References/QuestionDesign.md`" exists, so the two clearly-signaled one-level references are dangling and the content they promise (the actual workflow steps and the question-design patterns) is unreachable. Not anchor 1 (the SKILL.md itself is well-sectioned, not a wall of text) and not anchor 3 (references are well-signaled; the problem is they are broken, which defeats navigation entirely) — minimal functioning structure, so 2. | 2 / 5 |
Total | 12 / 20 Passed |