Stock & Industry Analysis (个股 & 行业 多维分析)

Overview

This skill turns a casual "帮我看看这只股票 / 这个行业" request into a rigorous, structured
report. It bundles two battle-tested prompt frameworks — one for individual stocks and one
for industries — and a repeatable data-gathering workflow. The agent fills the frameworks
with real, sourced data instead of generic filler, and always separates facts from judgment
while refraining from giving direct buy/sell advice.

When to Use

• A user asks to analyze a single listed company by name or ticker (e.g. "分析一下比亚迪 002594").
• A user asks to analyze an industry / sector / 赛道 (e.g. "分析一下光伏行业", "半导体板块怎么样").
• A user wants a peer comparison of companies inside an industry.
• A user uses trigger phrases: 个股分析, 行业分析, 股票诊断, 赛道分析, 板块分析.

Mode decision (read before gathering data)

• Individual stock mode — the request names a concrete company (often with a code). Use the
  8-section framework in references/individual-stock-framework.md. Its nested L2 already
  requires an industry read, so when deeper industry context is needed, run the industry
  framework first and fold the result into L2.
• Industry mode — the request names a sector with no single company focus. Use the 6-layer
  framework in references/industry-framework.md.
• If ambiguous, ask the user which they want. Default: a named listed company → 个股; a sector
  name only → 行业.

Workflow

Step 1 — Confirm the subject and mode

Capture the exact entity (company name + ticker, or industry name). If the ticker is missing
for a stock request, resolve it first (search the company name). State which framework will be
used so the user knows what to expect.

Step 2 — Gather data (deterministic, sourced)

Do not fabricate numbers. Pull from the following, in order of preference:

1. Structured market data — Prefer invoking the built-in finance skills
   (wb-finance-skill, westock-data, neodata-financial-search) for real-time prices,
   valuation (PE/PB/PEG), financials (revenue/profit/ROE/cash flow), and fund flows
   (北向资金, 融资融券, 主力资金). If they are unavailable, fall back to WebSearch/WebFetch
   against 东方财富 / 同花顺 / 雪球 / 巨潮资讯.
2. Industry research — Brokerage research reports, industry-association data, and
   government/statistical releases for market size, growth, policy, and cycle position.
   Use WebSearch/WebFetch; cite the source and date for each figure.
3. Sentiment & expectations — News flow, analyst consensus, and any visible expectation
   gaps (预期差) between market consensus and actual results.

For every quantitative claim, record its source and as-of date. If a figure cannot be
verified, say so explicitly rather than guessing.

Step 3 — Apply the framework

Load the matching reference file and follow its section structure exactly:

• Stock request → references/individual-stock-framework.md
• Industry request → references/industry-framework.md

Work section by section. Each section must contain a substantive analytical paragraph, not a
filled-in template blank. Use the embedded prompts verbatim as the outline; the operative
instruction set (结论先行, 同业对比表格, 事实vs判断, 不给买卖建议) is part of each framework file.

Step 4 — Personal-system overlay (stock mode only)

Section 七 of the stock framework asks to check buy/sell conditions against the user's personal
investment system. If the user's MA55 三层过滤量化策略 package is available in the current
workspace, apply its documented buy/sell signal rules to detect which signals are currently
triggered. Otherwise, describe qualitatively whether the current technical setup matches a
trend-following entry/exit without inventing a specific system. Never present this as advice —
frame it as "信号检测 / 条件自检".

Step 5 — Deliver

Present the full structured report. Lead every major section with its one-line conclusion.
Keep facts and judgment clearly distinguishable (e.g. label sub-points as 事实 / 判断 where
helpful). End with the 跟踪清单 / 催化剂-and-risk section so the user knows what to watch.

Output Conventions (always enforce)

• 结论先行: open with a one-line verdict (买入/持有/观望/卖出 for stocks; 推荐/中性/回避 for
  industries) plus the core reason.
• 同业对比 ≥ 5 家, rendered as a Markdown table (龙头对比 / peer comparison).
• 区分事实与判断: state verified data as fact, clearly mark interpretation as judgment.
• 不给买卖建议: only do fact-finding and signal detection; do not tell the user to buy or
  sell. Use phrasing like "当前触发了 X 信号" not "你应该买入".
• Cite data sources and as-of dates inline or in a short 数据来源 list at the end.

Resources

• references/individual-stock-framework.md — the complete 8-section individual-stock analysis
  prompt (multi-layer architecture, 三源验证, 资金面, 买卖点自检, 跟踪清单).
• references/industry-framework.md — the complete 6-layer industry analysis prompt
  (行业认知 → 宏观基础 → 周期位置 → 产业链 → 竞争结构 → 资金面与预期差).

These reference files are the canonical frameworks. Load the one matching the detected mode;
do not inline both into context unless the task spans both.