Content
71%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, actionable financial-analysis reference with concrete formulas, thresholds, scoring cards, and templates, plus embedded cross-check validation. Its main weakness is progressive disclosure: it is a long monolithic document with no split-out reference files.
Suggestions
Extract the 12-red-flag catalog and the output-format/DuPont templates into separate reference files (e.g. references/red-flags.md, references/output-template.md) and link to them from SKILL.md for one-level-deep navigation.
Show how the Python verification variables (net_income, cfo, accounts_receivable, revenue) are obtained from the data source mentioned in 注意事项 (tushare) to make the code fully copy-paste runnable.
Add an explicit numbered workflow with a validation checkpoint before producing the final output (e.g. run 勾稽验证 → only proceed if consistent → fill output template).
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is information-dense — ratio tables, the cash-flow quality matrix, and code/formula blocks convey a lot per token and assume accounting literacy — with only minor structural overhead (the one-line 概述 and repeated section framing). | 4 / 5 |
Actionability | It provides concrete, executable guidance: a Python 勾稽验证 block (accrual_ratio, receivable_growth, cash_change), thresholded scoring cards, and copy-paste 杜邦分析 and 输出格式 templates; the only gap is that input variables are not shown being loaded. | 4 / 5 |
Workflow Clarity | A clear analytical sequence (三表框架 → 勾稽验证 → 盈利质量 → 造假红旗 → 杜邦 → 输出格式) with embedded validation logic ('如果差异大 → 盈利质量存疑', 'abs(cash_change - cf_total) > 1 → 数据有问题') and decision-rule scoring thresholds; not a destructive/batch op so no cap applies. | 4 / 5 |
Progressive Disclosure | Section headers are well organized, but the ~260-line skill is monolithic with no external reference files — detailed catalogs (12 red flags, output templates, DuPont details) are inlined that could be split into one-level-deep references. | 3 / 5 |
Total | 15 / 20 Passed |