a-share-stock-dossier
v1.0.0Analyze A-share stocks and portfolios with analyst-grade, evidence-first reports. Use when the user asks for 个股分析、持仓复盘、逻辑是否还在、行业龙头、盘前/盘后策略、情绪+技术综合判断, especially when they want deep web verification and full process transparency (检索过程摘要 + 证据逐条分析) via web_search/web_fetch/browser plus Eastmoney/Tencen...
Installation
A-Share Stock Dossier
Overview
Produce professional analyst-style stock reports that are process-transparent, evidence-bound, and directly executable. Always split conclusions into two layers: - 产业逻辑(fundamental/industry logic) - 交易逻辑(price/flow/sentiment logic)
Default output mode is long-form process report unless user explicitly asks for a short summary.
Workflow
Step 1) Fix scope and objective
- Extract stock list, cost, position size, horizon (日内 / 次日 / 5-10日), and risk preference.
- Confirm output mode:
- 单票深挖
- 组合分层(A/B/C)
- 盘前执行单
- If screenshot is provided, parse first; ask only missing fields.
Step 2) Pull structured baseline before any narrative
Run:
python skills/a-share-stock-dossier/scripts/a_share_snapshot.py
--codes 603618,002149,002506,002475,002729,601116,601096
--with-indices --with-kline --kline-days 60 --pretty
Lock objective facts first: - Price/pct/range/volume/turnover - 5d/10d/20d return - MA5/MA10/MA20/MA60 context - Index mood and breadth proxy
Field reference:
- references/eastmoney-fields.md
Step 3) Run deep-search loop (before and during writing)
Use:
- references/source-checklist.md
- references/search-depth-protocol.md
Mandatory minimum evidence per stock: 1. Structured quote/kline data 2. One official source (CNINFO/exchange/company IR) 3. One mainstream finance source 4. One sector/leader verification source
Tool order:
- web_search discover
- web_fetch extract正文
- browser for JS-heavy/anti-bot/paginated/incomplete extraction
Step 4) Maintain retrieval log (hard requirement)
During analysis, build a step log S1..Sn.
Each step must include:
- 检索目标(why this search)
- 查询/页面(query/url)
- 摘要(1-3条关键事实)
- 来源等级(官方/主流媒体/社区)
- 对判断影响(supports/weakens/conflicts)
If a conclusion appears without supporting steps, do not keep it in final recommendations.
Step 5) Write stock analysis in fixed order
For each stock, output strictly in this order: 1. 公司业务与收入/应用场景定位 2. 当前市场叙事与叙事阶段(启动/强化/分歧/退潮) 3. 行业龙头与板块阶段(强度/轮动/分化) 4. 技术面(趋势、关键位、失效位) 5. 舆情与事件(利多/利空/争议) 6. 双逻辑判断 - 产业逻辑:在 / 弱化 / 失效 - 交易逻辑:在 / 弱化 / 失效 7. 明日三情景(强/中/弱)触发条件 -> 动作 8. 证据绑定(E1/E2/E3/E4)+ 置信度
Use template:
- references/report-template.md
Step 6) Continuous-search triggers during writing
Pause and re-search immediately if: - only one source supports a key claim - key event is stale (>7 days) and no update is checked - price/volume behavior conflicts with narrative - wording becomes uncertain(可能/大概/据说) - sector leader list mismatches same-day board behavior
Step 7) Conflict resolution and stop rule
- Unify basis first (timestamp, adj/non-adj, intraday/close)
- Priority: official > exchange data > mainstream media > community
- If unresolved, keep explicit uncertainty notes
Stop searching only when: 1. each core conclusion has >=2 sources and >=1 official/preferred source 2. recent two re-search rounds add no high-value facts 3. conflicts are either resolved or explicitly marked
Step 8) Portfolio decision + self-correction
After all stocks: - Rank A/B/C: - A: 产业逻辑与交易逻辑同向 - B: 产业逻辑在、交易逻辑弱 - C: 交易逻辑受损 - Give one-line portfolio action (cut/hold/wait + why) - Add self-correction: - 2-3 weak points in this round - how to recalibrate thresholds next round
Output requirements (must follow)
Default to detailed analyst-report style with this top-level structure: 1. 检索过程纪要(S1..Sn) 2. 市场底色(结构化数据) 3. 逐股深度分析(证据逐条绑定) 4. 组合分层与执行重点 5. 本轮不确定性与下轮修正计划
Never output only short conclusions unless user asks explicitly.