Content
63%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, knowledgeable reference with strong executable code and China-market specifics Claude would not know, delivered as a monolithic 890-line file. Its main weaknesses are padding from textbook strategy explanations and the total absence of progressive disclosure via separate reference files.
Suggestions
Split the file: move the product/index classification tables (§1, §5.4) and the Tushare data-access code templates (§7) into references/ and scripts/ files, keeping SKILL.md as a concise overview with one-level-deep links.
Trim or cut sections explaining widely known concepts (core-satellite, volatility decay derivation, 60/40 and all-weather portfolios) down to the China-specific parameters that actually add value.
Fix the data contract inconsistency (get_etf_nav returns trade_date/nav/accum_nav but downstream code uses unit_nav and fund_share) and add explicit handling for empty or failed Tushare API responses.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The China-specific reference data (ETF tickers, fee levels, tax rules, QDII premium thresholds, index-adjustment schedule) is genuinely non-obvious and earns its tokens, but roughly a third of the ~890-line file explains concepts Claude already knows (core-satellite strategy, volatility decay math, 60/40 portfolios, single-factor regression). The file could be tightened considerably without losing value. | 3 / 5 |
Actionability | Abundant concrete guidance: executable, docstringed Python (etf_score, calc_tracking_error, monitor_qdii_premium, fund-flow analysis), a 5-step selection procedure, and numeric thresholds throughout. Minor gaps keep it from a 5: the code depends on project-internal modules (src.data_gateway) absent from the bundle, and etf_fund_flow_analysis consumes unit_nav/fund_share columns that get_etf_nav does not return. | 4 / 5 |
Workflow Clarity | Multi-step processes are clearly sequenced (the Step 1-5 comparison framework in §3.1, scenario templates in §8, risk-graded monitoring logic in §7.3), and the analytical read-only nature means the destructive-operation cap does not apply. Missing explicit error-recovery or validation checkpoints for data-access failures (empty API responses, rate limits) prevents a 5. | 4 / 5 |
Progressive Disclosure | Internal structure is good (nine numbered sections with headers and tables), but it is a monolithic ~890-line SKILL.md with no bundle files at all — the product classification tables, index-system reference, and data-access code templates are exactly the material that belongs in references/ and scripts/ files linked one level deep. | 3 / 5 |
Total | 14 / 20 Passed |