Content
65%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 strong, information-dense domain skill with executable data-access code, concrete numeric thresholds, and a clear output contract. The main gaps are the lack of an explicit step-by-step workflow with validation checkpoints and the absence of progressive disclosure — everything lives inline in one long file.
Suggestions
Add an explicit numbered workflow (fetch flows → compute signals → score composite → render output) with a validation step (e.g., cross-check daily Southbound against HKEX before signaling) to anchor execution order.
Complete the composite scoring logic: the connect_score dict is all zeros with no function mapping raw flows/percentiles to the -2..+2 component scores.
Split the sector allocation tables and signal frameworks into a references/ file (e.g., references/signal-frameworks.md) and link them one level deep, keeping SKILL.md as a lean overview plus data access and output format.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly lean and dense with domain-specific judgment Claude cannot derive (quota thresholds, sector weight ranges, signal cutoffs, tushare field pitfalls), with only minor over-explanation — e.g., the 'Why Northbound matters' bullets and 'smart money' framing could be tightened. Fits 'efficient; minor instances of over-explanation that could be trimmed'. | 4 / 5 |
Actionability | Mostly executable guidance: real API calls with parameters and column documentation (moneyflow_hsgt, hsgt_top10, get_southbound_flow, yahoo_download), concrete numeric signal functions, and a fill-in output template. Minor gaps keep it below 5 — the composite 'connect_score' dict is left initialized to zeros with no scoring function showing how to compute each dimension, and 'ah_discount': [] is an unexplained placeholder. | 4 / 5 |
Workflow Clarity | The sequence (understand signals → fetch data → compute composite → render output format) is present but only implicitly through section order; there is no explicit step list or validation checkpoints. The tushare cumulative-field warning and Eastmoney↔HKEX cross-verification note act as partial data validation, fitting 'steps listed but validation gaps; sequence present but checkpoints missing or implicit'. | 3 / 5 |
Progressive Disclosure | Sections are well-organized with clear headers, but the ~260-line body is monolithic with no bundle files — the signal frameworks, sector allocation tables, and output template are natural candidates for one-level-deep reference files. Fits 'some structure but could be better organized' rather than the appropriately-split structure of a 4. | 3 / 5 |
Total | 14 / 20 Passed |