A股新闻研报检索

当用户需要获取与时间相关或基于来源信息的金融新闻、公告、研究报告、政策条文、交易规则、市场事件说明或影响分析时,可以使用此技能通过 Eastmoney MX 搜索来获取特定领域的金融信息。相比一般的网络搜索,这种方式能更精准地获取与中国股市相关的金融信息。不过,不要使用该功能来查询精确的报价/财务报表数据(建议使用 mx-data 功能)、基于条件的股票筛选(建议使用 mx-xuangu 功能)、设置观察名单的操作(建议使用 mx-zixuan 功能),以及模拟交易操作(建议使用 mx-moni 功能)

Install

openclaw skills install @zoeluli7459-dev/mx-search

mx-search

Use mx_search.py to retrieve finance news and event context from Eastmoney MX selected sources.

Trigger Boundaries

  • Use for "latest", "today", "recent", "why did it move", announcements, research reports, policy updates, trading rules, event timelines, and source-backed market explanations.
  • Use before analysis when the user asks about a specific finance event whose facts may have changed.
  • Do not use for pure numerical quote/financial data; call mx-data.
  • Do not use for stock universe filtering; call mx-xuangu.

Inputs

  • Required: event/topic/entity plus what kind of source is needed if known, such as announcement, research report, policy, or news.
  • Ask a follow-up only when missing entity/time/source type would likely retrieve the wrong event.
  • Require MX_APIKEY. Optional: set MX_OUTPUT_DIR; otherwise output goes to ~/.codex/skills-output/mx_data/output.

Workflow

  1. Rewrite the user request into a search query with entity, event, date window if known, and source preference.
  2. Run:
bash
python /Users/lu/.codex/skills/mx-search/mx_search.py "贵州茅台最新研报 机构观点"
  1. Read terminal results for title, source, date, and content trunk.
  2. Load the generated .txt for longer extracted content or .json when metadata matters.
  3. If source ranking or result fields are unclear, read references/result-fields.md.

Output Contract

  • Cite source names and dates from the returned results.
  • Separate facts, source claims, and your inference.
  • For impact analysis, state the transmission logic and uncertainty instead of presenting source snippets as certainty.
  • Include generated file paths when useful for audit.

Failure Handling

  • Missing MX_APIKEY: stop and ask the user to configure it.
  • Sparse results: broaden the query or try the official name/code of the entity.
  • Conflicting sources: present the conflict and prefer official announcements or exchange/company disclosures where available.
  • API limit/auth/network errors: report the exact issue and do not fill gaps from memory.

Validation

  • Use evals/evals.json after changing routing or output behavior.
  • When changing the description, add both positive trigger examples and near-miss negative examples.