Content
81%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 highly actionable, well-sequenced decision procedure with concrete scoring rules, formulas, degradation paths, and a delivery-gating checklist — excellent on actionability and workflow clarity. Its weakness is duplication: the inline output template and search-query lists restate content already in the bundle's reference files, inflating the body to ~350 lines when a leaner overview pointing to those files would serve better.
Suggestions
Replace the ~80-line inline markdown template in 步骤 6 with the required-section list plus a pointer to `references/output-template.md` (e.g. "章节顺序固定:核心结论/五维分析/价格区间/交易计划/风险/术语/来源;骨架直接复制 `references/output-template.md`"), since the reference file already bills itself as the copyable skeleton.
Trim the P0/P1/P2 query blocks to just the priority rule and 2-3 example queries, delegating the full situational keyword combinations to `references/search-queries.md`.
Move the worked example in 步骤 6 (the filled 五维加权得分 table and example values) into the output template reference, keeping only the empty schema in the body.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is domain-specific rather than explaining known concepts, but ~80 lines of the full markdown output template are inlined in 步骤 6 even though `references/output-template.md` (325 lines, "可直接复制的 markdown 骨架") exists for exactly that, and the P0/P1/P2 query blocks duplicate `references/search-queries.md`. This is the 'mostly efficient but could be tightened' anchor; not 4 because the duplication of two existing reference files is a substantial, avoidable token cost, and not 2 because the material is skill-specific guidance rather than padded concept explanation. | 3 / 5 |
Actionability | Everything is executable: literal search query strings ("{ticker} earnings after-hours", "ES NQ futures overnight"), a -2..+2 scoring scale with worked examples and weight thresholds (> +1.0 = 强看多), a probability formula with explicit clamp rules (35%-65%, hard clamp at 65%), a position-sizing formula ("仓位(股数) = (账户净值 × 风险预算%) ÷ (入场价 − 止损价)"), fixed file naming, and a concrete self-check checklist. Copy-paste-ready guidance covering common cases; nothing pseudocode. | 5 / 5 |
Workflow Clarity | A 7-step strictly ordered flow (标的解析 → 搜证 → 打分 → 推演 → 交易计划 → 写文件 → 摘要) with explicit error-recovery paths: search failure downgrades confidence and mandates an "信息缺失" annotation, hedge signals force a confidence downgrade, and the 自检清单 gates delivery ("未通过任何一项 = 重新写,不能交付") — clear sequence with validation checkpoints and feedback loops. | 5 / 5 |
Progressive Disclosure | Four real, one-level-deep references (five-dimension-framework.md, terminology.md, output-template.md, search-queries.md — all verified on disk with substantive content), each signposted inline ("详细规则见 references/five-dimension-framework.md") and summarized in the 资源引用 section with what each contains. Not 5: the body inlines the full output template and full query lists whose canonical home is those reference files, so the split between overview and detail isn't clean. | 4 / 5 |
Total | 17 / 20 Passed |