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 dense, well-structured A股 earnings-surprise playbook with genuinely concrete formulas, thresholds, calendar timing, and an output template — its main weaknesses are the absence of inline validation checkpoints in the four-step workflow and a monolithic single-file layout where per-method detail (each of the five core concepts) would sit better in one-level-deep reference files. No bundle files exist, so progressive disclosure rests entirely on the in-file organization.
Suggestions
Insert explicit validation checkpoints into the four-step framework, e.g. after step 2 verify '覆盖分析师 ≥ 3 家且历史偏差 ≥ 8 个季度，否则降级为定性判断' before computing SUE.
Split each 核心概念 (Top-Down, Bottom-Up, SUE, PEAD, 预期修正动量) into one-level-deep reference files (e.g. references/pead.md) and keep SKILL.md as a concise overview with clearly signaled pointers.
Replace the comment-style pseudocode blocks with runnable snippets (data retrieval + SUE computation) or explicitly justify the formula-as-pseudocode style, to close the gap toward fully copy-paste-ready guidance.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and operational — tables, formulas, thresholds, and a config dict with almost no filler prose, and the finance content it does explain (A股 PEAD holding windows, disclosure calendar, short-selling constraints) is market-specific rather than what Claude already knows. It is not a 5 because sections like the 白酒/贵州茅台 top-down worked example and the generic profit-margin bullet points could be trimmed. | 4 / 5 |
Actionability | Concrete, executable-level guidance dominates: exact formulas (SUE, ERM, dispersion with computational definitions), explicit thresholds (SUE > +1.5, hold 40 trading days, stop_loss -0.08), a complete portfolio config dict, an A股 disclosure calendar, and a fill-in output template. It stops short of 5 because the Python blocks are annotated formulas/pseudocode rather than copy-paste runnable code (no data-retrieval or computation pipeline), though as an instruction-oriented analysis skill the per-scoring-note absence of runnable code is only a minor gap. | 4 / 5 |
Workflow Clarity | The '盈利分析四步法' gives a clear ordered sequence (构建预测 → 获取一致预期 → 计算偏差 → 信号生成) reinforced by the calendar and output template, matching the anchor 'steps listed but validation gaps; checkpoints missing or implicit'. Preconditions such as 'SUE计算需要至少8个季度的历史预测偏差数据' exist only in the trailing 注意事项 rather than as inline validation checkpoints, and there are no feedback loops (e.g., re-check dispersion or coverage before acting on a signal) — so it does not reach 4, though the sequence is coherent enough to stay above 2. | 3 / 5 |
Progressive Disclosure | There is good in-file structure (clear 概述/核心概念/分析框架/输出格式/注意事项 headers), but the skill is ~195 lines with five substantial method sections (Top-Down, Bottom-Up, SUE, PEAD, revision momentum) all inlined in a monolithic SKILL.md with no references/, scripts/, or assets/ files and no pointers to split-out detail. This matches 'some structure but content that should be separate is inline'; it is above 2 because section organization and navigation within the file are genuinely clear. | 3 / 5 |
Total | 14 / 20 Passed |