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 delivers strong domain substance — A-share-specific clause thresholds, a decision matrix, concrete screening criteria, and an output template — but it is a monolithic, explanation-heavy document. The strategy engine is unexecuted pseudocode, workflows lack validation checkpoints, and everything lives in one file rather than being progressively disclosed.
Suggestions
Remove the duplicate 转股价值 formula (keep one occurrence) and cut textbook explanations of basic mechanics (纯债价值含义 bullets), keeping only A-share-specific thresholds and conventions Claude would not know.
Make the strategy layer executable or delete the stub: either implement ConvertibleBondEngine.generate (double-low computation, Top-N selection, monthly rebalance) with real tushare field names, or drop the class and specify the signal rule in prose with concrete numbers.
Add validation checkpoints to the rotation/backtest workflow (e.g., verify 转股价/强赎公告 fields are present and non-null after fetching, cross-check 双低值 against price and premium before ranking).
Split detail into one-level-deep reference files (e.g., references/clause-playbooks.md for 下修/强赎/回售 game details, references/backtest.md for the 回测框架) and keep SKILL.md as a concise overview with clearly signaled links.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and mostly signal (clause trigger thresholds, decision matrix, output template), but it re-explains textbook convertible-bond mechanics Claude already knows — e.g., the 转股价值 formula appears twice (lines 31-32 and 72) and bullets like "纯债价值越高 → 债底保护越强 → 下跌空间有限" are explanatory padding. Not a 2 because there is no generic library/tutorial padding and most sections carry A-share-specific detail Claude cannot infer. | 3 / 5 |
Actionability | There is one complete executable function (bond_floor) plus concrete numeric thresholds (双低筛选: <130, 价格<115, 溢价率<30%, 评级≥AA-) and a concrete output-format template. However, the core strategy engine (ConvertibleBondEngine) is a pseudocode stub with 'pass', and data retrieval via tushare is mentioned without any executable code — key details are missing. | 3 / 5 |
Workflow Clarity | The 轮动流程 gives a real 5-step sequence (筛选→打分→排名→选Top 15-20→月调仓) and 下修/强赎/回售 analyses are ordered, so a sequence exists. But there are no validation checkpoints anywhere (no data-quality checks, no signal sanity checks), and the monthly batch rebalancing/backtest workflow falls under the rubric's batch-operation cap of 3. Not a 2 because step 1 does embed an implicit filter (剔除已退市/已公告强赎/评级<A+) and the sequence is coherent and well-defined. | 3 / 5 |
Progressive Disclosure | The file is well-sectioned (概述/基础概念/三维估值/条款博弈/双低策略/轮动/输出格式/注意事项) with no nested or dead references, but it is a single ~250-line monolith: the detailed clause-game playbooks and the backtest framework are inline content that would fit better in separate reference files. There are no bundle files at all, so all detail must load into context on every invocation. Not a 2 because the structure and section headers make it genuinely navigable; not a 4+ because nothing is split out and the file exceeds the size where a single file can score 5. | 3 / 5 |
Total | 12 / 20 Passed |