Content
68%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 tight, executable, and well-organized, with a complete OKX retrieval example and runnable indicator templates. Main gaps are the absent retrieval examples for Tushare/yfinance, implicit rather than sequenced workflows, and minor redundancy around the 300-row pagination limit.
Suggestions
Add short retrieval snippets for Tushare and yfinance (or a pointer to where their call patterns differ), since both are listed as supported sources but only OKX has code.
State the 300-row pagination limit once (the Pitfalls section) and remove the duplicates in the code comment and Parameters table.
Add an explicit validation checkpoint after data retrieval, e.g. "If the returned DataFrame is empty (common with Tushare score < 2000), stop and report rather than running the backtest."
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean with no explanation of concepts Claude already knows, but "at most 300 rows per request" appears three times (code comment, Parameters table, Pitfalls) and the Parameters table largely repeats the code-sample comments. This matches anchor 4 (efficient with minor instances that could be trimmed) rather than anchor 5's every-token-earns-its-place. | 4 / 5 |
Actionability | The OKX snippet is fully executable (imports, params, DataFrame construction, dtype conversion) and the indicator templates and config JSON are copy-paste ready, matching anchor 4. It falls short of anchor 5 because Tushare and yfinance are listed as supported sources but have no retrieval code, leaving the common cases for those sources uncovered. | 4 / 5 |
Workflow Clarity | A rough sequence exists (set interval in config.json, run backtest tool; or install deps, call API, compute indicators) but it is implicit across sections with no explicit checkpoints — e.g. the Tushare empty-data caveat is a pitfall note rather than a verify-data step. This fits anchor 3 (sequence present but checkpoints missing or implicit); it exceeds anchor 2 because the steps are well defined, and does not reach anchor 4's most-checkpoints-present. | 3 / 5 |
Progressive Disclosure | A single well-organized ~96-line file with clear sections (Purpose, Backtest Configuration, Data Sources, API, Indicators, Parameters, Pitfalls, Dependencies) and no buried or nested references, matching anchor 4. It misses anchor 5 because it is over the simple-skill threshold and all API/indicator detail is inline where a reference file split could apply. | 4 / 5 |
Total | 15 / 20 Passed |