Content
72%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The body is concise, well-structured, and self-contained with a useful pre-delivery quality gate, but its guidance is more checklist than executable methodology and the research workflow is organized by topic rather than as an explicitly sequenced procedure with feedback loops.
Suggestions
Add a concrete research workflow as an ordered sequence (e.g. 1. frame the decision question → 2. gather per relevant mode → 3. synthesize into the output format → 4. run the Quality Gate → 5. fix any failing item and re-check), so the process is a tracked procedure rather than topical sections.
Strengthen actionability with executable specifics: name concrete data sources or source types for each mode and give a worked TAM/SAM/SOM example with the explicit assumptions the standards require, instead of only naming the top-down/bottom-up methods.
Turn the Quality Gate into an explicit feedback loop — state what to do when an item fails (revise the finding, re-source the number, or flag the caveat) before re-validating, rather than only listing pass criteria.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is a lean set of checklists and structural guidance (Research Standards, Modes, Output Format, Quality Gate) with a high-signal opener ('生成支持决策的研究，而非流于形式的研究报告') and no explanation of concepts Claude already knows, matching the level-3 'lean and efficient, every token earns its place' anchor. | 3 / 3 |
Actionability | It offers concrete collection lists and a numbered output structure, but stays at the checklist level — e.g. '来自报告或公开数据集的自上而下估算' names a method without specific data sources, TAM/SAM formulas, or a worked example, fitting level-2 'some concrete guidance but incomplete, missing key details' rather than fully executable level-3. | 2 / 3 |
Workflow Clarity | Content is organized by topic with an explicit pre-delivery Quality Gate checkpoint, but the research process is not laid out as a crisp ordered sequence (collect → synthesize → validate → deliver) and lacks a validate→fix→retry feedback loop, placing it at level-2 'steps present but checkpoints implicit / sequence gaps' rather than level-3. | 2 / 3 |
Progressive Disclosure | With no bundle files present and ~70 well-organized lines under clear headers (何时激活, 研究标准, 常用研究模式, 输出格式, 质量门禁), the skill is appropriately self-contained and easy to navigate, meeting the simple-skills allowance for level 3 without external references. | 3 / 3 |
Total | 10 / 12 Passed |