Content
57%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 dense, well-organized reference rich in China-market specifics and mostly executable code, but it is a monolithic wall of reference material with no progressive disclosure into separate files, no validation checkpoints in its workflows, and noticeable padding from redundant concept explanations and docstrings.
Suggestions
Split the monolithic SKILL.md into an overview plus one-level-deep reference files (e.g. references/products.md for the product/index catalogs, references/data-api.md for the tushare code, references/strategies.md for sections 4 and 6), keeping SKILL.md a lean map with pointers.
Add validation checkpoints to the analysis workflows — e.g. verify data alignment/completeness before computing tracking error, sanity-check premium_pct against IOPV freshness for QDII ETFs, and confirm scale data exists before ranking ETFs.
Trim content Claude already knows: replace the tracking-error/IR/volatility-decay derivations and verbose per-function docstrings with brief formulas and one-line usage comments, keeping only the China-specific thresholds and codes.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | China-specific data (fund codes like 510300, fee tables, QDII premium thresholds, tax rules) earns its place, but the ~870-line body re-explains concepts Claude already knows (tracking error mechanics, core-satellite strategy, volatility decay derivation) and pads nearly every code block with multi-line docstrings. Mostly efficient with some unnecessary explanation, fitting anchor 3. | 3 / 5 |
Actionability | Most code is executable and complete (get_etf_list, calc_tracking_error, monitor_qdii_premium with real tushare API calls), but several snippets are fragments assuming undefined context (effective_spread/impact_cost lines, the z-score pair-trading block, the comment-only fee_drag_analysis example). Anchor 4 fits better than 5 due to these gaps. | 4 / 5 |
Workflow Clarity | A clear selection sequence exists ("Step 1: 规模筛选 → ... Step 5: 基金公司") plus scenario templates in section 8, but there are no validation checkpoints anywhere — no data-quality checks before computing tracking error, no sanity checks on results. Sequence present, checkpoints missing, matching anchor 3. | 3 / 5 |
Progressive Disclosure | Sections are well-organized with clear numbered headers, but this is a monolithic ~870-line SKILL.md with product-catalog tables, a factor table, and a full Tushare API reference inlined — content that clearly belongs in separate reference files. No bundle files exist at all, so everything is inline; anchor 3 fits (structure present, content that should be separate is inline). | 3 / 5 |
Total | 13 / 20 Passed |