Content
63%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 lays out a clear, reasonably actionable extraction workflow with good failure handling, but is diluted by generic constraint boilerplate and squanders its one reference file by never linking to it. Fixing the orphaned reference and trimming the common-constraints block would lift both conciseness and progressive disclosure.
Suggestions
Reference the existing references/prompt_templates.md from the body (e.g. under 语义提取: 'prompt 模板见 [prompt_templates.md](references/prompt_templates.md)') so its templates and per-input-type guidance are actually discoverable.
Add a concrete example of the structured output (one short text or yaml block showing 主题/关键词/核心问题) so the required format is unambiguous without inference.
Trim or externalize the 公共硬约束 and bensz-collect-bugs boilerplate that is unrelated to topic extraction, keeping only the skill-specific constraints inline.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The core flow sections (识别输入类型, 提取内容, 语义提取, 校验, 失败与恢复) are lean bullet lists, but a large share of the body is generic infrastructure boilerplate — the bensz-collect-bugs procedure and the 8-bullet 公共硬约束 block — that does not advance topic extraction and could be tightened or externalized, matching 'mostly efficient but includes some unnecessary explanation'. | 3 / 5 |
Actionability | Guidance is concrete for an instruction-only skill: per-input-type handling ("图片：依赖 LLM 原生视觉能力", "文件夹：递归扫描并合并 .md/.txt/.pdf"), numeric output specs ("5-10 个英文标准术语", "2-5 个具体研究问题"), and per-failure recovery actions. Not 5 because the body never shows a concrete example of the text/yaml/json output the skill must produce, leaving the format specification to inference. | 4 / 5 |
Workflow Clarity | A clear sequence runs 输入 → 识别输入类型 → 提取内容 → 语义提取 → 格式化 → 输出, with a 校验 checkpoint and a dedicated 失败与恢复 section covering five failure modes. Not 5 because validation is a static checklist with no explicit feedback loop (e.g. 'if 校验 fails, regenerate') tying it back into re-extraction. | 4 / 5 |
Progressive Disclosure | The bundle contains references/prompt_templates.md with genuinely useful prompt templates and per-input-type guidance, but the body never mentions or links it — the reference is orphaned, so Claude would never load it and its content goes unused. This matches 'references present but not clearly signaled'; not 2 because the body itself is well-sectioned and short enough to stand alone. | 3 / 5 |
Total | 14 / 20 Passed |