Content
50%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 is a well-organized, information-dense reference whose specialized regulatory data (thresholds, fee schedules, tax tiers, dated rule changes) is exactly what Claude cannot be trusted to recall. Its weaknesses are structural: ~300 lines all inlined in SKILL.md with no progressive disclosure, non-executable pseudocode where runnable backtest-constraint code would fit, and no validation checkpoint for the analysis workflow it prescribes.
Suggestions
Split the four per-market sections (A股/港股/美股/加密) into references/ files (e.g., references/a-share.md, references/us.md) and keep SKILL.md as an overview plus the 约束矩阵 and output format — the file currently inlines ~295 lines of reference detail with no external files.
Replace the 涨跌停/T+1 pseudocode blocks with an executable pandas snippet a backtest engine can actually call, or drop the unused 'pip install pandas numpy' dependency section if no runnable code is intended.
Add a validation checkpoint to the 分析框架 workflow, e.g. 'verify the backtest reproduces historical limit-up non-fills on a known date range before trusting strategy returns', so the prescribed compliance analysis has a feedback loop.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The dense tabular format carries genuinely specialized, time-sensitive data (维持担保比例 thresholds, 2023 印花税减半, dividend-tax holding tiers) that earns its tokens. But the 约束矩阵 repeats material already covered in 核心概念, and sections like 美股交易时间 and the 7%/13%/20% circuit-breaker levels are basics Claude already knows. This is 'mostly efficient but includes some unnecessary explanation or could be tightened', not the noticeably-padded level 2. | 3 / 5 |
Actionability | There is real concrete guidance — the 跨市场合规检查清单, exact A股 cost breakdown (佣金 0.025%, 印花税 0.05%), and a filled-in output-report template — but the only code is non-executable pseudocode ('if daily_return >= limit_up: buy_signal无法执行, 推迟到次日'), and the 'pip install pandas numpy' dependency is used by nothing in the skill. Matches 'some concrete guidance but incomplete; pseudocode instead of executable code'. | 3 / 5 |
Workflow Clarity | The 分析框架 section provides a numbered sequence (约束矩阵 → 合规检查清单 → 税务影响) and a concrete output format, giving an analyzable path. However, checkpoints are missing: nothing tells the reader how to verify a backtest correctly encodes a rule (e.g., comparing simulated fills against limit-up days), so it sits at 'steps listed but validation gaps'. This is an analysis/reference skill, not a destructive or batch operation, so no hard cap applies. | 3 / 5 |
Progressive Disclosure | Sections are well-organized with clear ##/### headers, but the file is ~295 lines of per-market reference detail (A股, 港股, 美股, 加密 each ~40-60 lines) with zero external reference files — content that naturally belongs in per-market reference files is fully inlined. This is over the 50-line threshold where organized-sections-alone would justify a 5, matching 'some structure but could be better organized; content that should be separate is inline'. | 3 / 5 |
Total | 12 / 20 Passed |