ClawHub

Capital Market Report

Generate high-signal, impact-driven capital market anomaly and rumor reports. Focuses on actionable business signals, expectation-breaking news, and deep logical deduction rather than routine index reading. Uses local scanners to digest Chinese and global financial media, Reddit, and Twitter. Features strict market-isolation rules to prevent cross-border misattribution and requires real source URLs.

Audits

Pending
ClawScanPass

Agentic behavior and permission review.

Static analysisPass

Pattern checks against bundled files.

VirusTotalPending

Multi-engine malware detections and file reputation.

Install

openclaw skills install capital-market-report

Capital Market Report (High-Signal Anomaly Edition)

Generates forward-looking business deduction reports based on an "Absolute Impact Threshold." This skill abandons traditional macroeconomic index reading (e.g., "Nasdaq down 1%"), shifting instead to deep scraping of domestic and international forums, news, and social media to lock in supply-chain anomalies and earnings explosion points with extreme expectation gaps.

Core Philosophy (Absolute Impact Threshold)

  1. Zero Routine Data: No longer reports routine market data like index points or daily percentage changes. Focuses entirely on nuclear-level anomalies in the business world.
  2. Dynamic Capacity: Abolishes mechanical rules like "must have 3-5 items." If only 2 events meet the threshold today, report 2; if 8 major global supply-chain-shaking news break, report all 8.
  3. Core on Selected Assets, Inclusive of Global Leaders: While deeply scanning specific target assets (e.g., Chinese ADRs, A/HK tech stocks, AI/consumer/EV supply chains), it absolutely must not miss strategic turning points from global tech giants (e.g., the "Magnificent Seven" or core global AI leaders).

Execution Pipeline & Toolchain

Step 1: Launch Underlying Scanners

You must run the following information scraping tools:

1. Chinese Financial Core News Scraper (Scraping domestic sources like Cailianshe, Wall Street CN, Sina Finance):

cd ~/.openclaw/skills/capital-market-report; uv run scripts/news-processor.py --delta --delta

2. Overseas Anomaly & Rumor Radar (Monitoring Reddit WSB, CoinGecko, Yahoo Finance movers, Google News rumors): (Note: Requires the stock-analysis skill to be installed)

uv run ~/.openclaw/skills/stock-analysis/scripts/hot_scanner.py
uv run ~/.openclaw/skills/stock-analysis/scripts/rumor_scanner.py

Step 2: Comprehensive Inclusion of Major Market Events

From the scraped results, retain news based on the following rules:

  • Include All High-Attention Events: Any news heavily discussed by the market (e.g., major AI model releases like Claude/GPT updates, big tech earnings, macro data, geopolitical shifts) MUST be included, even if they perfectly meet market expectations. Do not filter out highly focused topics.
  • Retain Anomalies: Keep extremely strong earnings reversals, supply-chain-level product delays, or major rumors.
  • Red Line: Do not aggressively filter out major news just because it lacks a "shock" factor. If the market cares, it goes in the report.
  • Freshness Mandatory (24h): Every news article MUST have been published within the last 24 hours from the report generation time. Check the published date before including any article. If an article is older than 24 hours, discard it immediately — it has already appeared or should have appeared in a previous report. This prevents stale news from showing up as if it were current.
  • 财报事件强制识别（2026-05-13 Phil 指令）：新闻标题中包含「一季/二季/三季/四季/全年」+「营收/净利润/同比增长/同比下降/毛利率/净亏损」等关键词的，无论外面包了什么壳（如"电报解读""机构点评"等），一律强制识别为财报事件，纳入简报。不得因标题包装而过滤。
  • 市场影响力判断原则（2026-05-13 Phil 指令）：公司重要性不应依赖固定白名单，而应基于事件本身的市场冲击力做主观判断。巨头公司（如腾讯、阿里巴巴、NVDA 等）的重大事件天然有市场影响力；但也不应忽略中小公司的行业拐点信号。判断标准是事件本身能否影响市场预期，而非公司的市值标签。

Step 3: Mandatory Source Tracking (Real URL Verification)

Red Line: Every piece of news reported must include a real source URL [Read Original](URL).

  • The URL is natively extracted and provided by the underlying news-processor.py script from the original RSS or HTML feeds.
  • Do not invent URLs or use generic domain homepages. Rely on the exact link returned by the scripts.

Step 4: Isolated Deduction & Localization

Perform strict logical deduction on the selected events:

  • 推测性内容红线（绝对禁止）: 任何涉及具体公司立场、态度、游说方向、战略意图的判断，必须有原文明确依据。如果原文没有提及，直接删除该条点评，不要猜测。禁止用"可能"、"倾向于"、"说明"等词包装无原文依据的推断。宁可不写，不可编造。
  • Market Isolation Red Line: Rigorously distinguish the "country/market" where the anomaly occurred. For example, a surge in US domestic airfares can only be deduced as bullish for US airlines and US OTAs; it absolutely cannot be forcefully applied to Chinese companies like Trip.com.
  • If the event is a shock to an overseas giant (e.g., Honda taking a massive loss), the deduction must clarify whether the logical link to its global competitors genuinely holds up.
  • Language Localization: Although this skill description is in English, the final report generated MUST be written in the user's primary conversational language (e.g., if the user communicates in Chinese, the report must be in Chinese).

Step 5: Rolling Updates & 24-Hour Delta Extraction (Temporal Event Tracking)

For multi-source concurrency or rolling reports on the same market event, DO NOT simply discard duplicate news items. Instead, apply strict "Delta Extraction" tracking based on all reports from the past 24 hours:

  • Definition of Delta: "Delta" means new information not covered in ANY report from the past 24 hours (temporal delta), NOT whether the news broke market expectations.
  • Baseline Comparison: You MUST load ALL reports from the past 24 hours before generating the new report. Reports are stored in ~/.openclaw/workspace-group/memory/capital_market_report_*.md (timestamped filename pattern capital_market_report_YYYYMMDD_HHMM.md). Read every report whose timestamp is within 24 hours of now, and build a merged set of all events/topics they covered.
    • This prevents the scenario where news published at 08:00 appears in the 09:00 report, then gets re-published at 14:00 and falsely flagged as "new" in the 15:00 report (because the 15:00 report only compared against 12:00).
    • By checking against ALL 24h reports, any topic that appeared at any point in the past day is recognized as existing coverage.
  • Extract Delta (New Information): Specifically pull out any new data, new official statements, or new market reactions that weren't in ANY previous report from the past 24 hours.
  • Visual Labeling: Use tiered labeling for event tracking updates. For example:
    • 🔴 [增量更新 - 关键细节/市场反应] for news containing substantial new facts since any prior coverage.
    • ⚪ [跟进报道 - 与已有报道相比无新增] for news that merely repeats facts already covered in past 24h reports.
  • Save State: After generating the new report, save it to ~/.openclaw/workspace-group/memory/capital_market_report_YYYYMMDD_HHMM.md (using the report's generation time). Then DELETE any capital_market_report_*.md files older than 24 hours to keep the directory clean. Also update memory/last_capital_market_report.md as a convenience symlink or copy for backwards compatibility.

飞书排版规则（强制）

  • 表格谨慎使用：飞书手机端单元格超过 8 个汉字会强制换行，阅读体验差
  • 判定规则：使用表格前检查，如果任一单元格内容超过 8 个汉字 → 改用列表、粗体、emoji
  • 如果所有单元格都很短（纯数字、短词如"利空/利好"、公司名 ≤ 4 字）→ 表格可用
  • 多组数据对比优先用列表嵌套或分段展示

🔴 防幻觉强制规则（Anti-Hallucination Guardrails）

每次生成报告前必须逐条检查，违反任何一条 = 报告作废。

规则1：时间校验（Temporal Verification）

  • 每一条涉及公司财报、盘后涨跌、经济数据的新闻，必须确认该事件是否已在当前北京时间发生。
  • 盘后数据规则：北京时间 06:00 之前（美东 18:00 之前），不存在当日美股盘后数据。北京时间 21:00 之前，不存在当日美股收盘数据。
  • 如果盘后数据尚未产生，绝对不写盘后涨跌幅。

规则2：财年 vs 日历年（Fiscal vs Calendar Year）

  • 警惕标题中的 "Q1 2026"——部分公司（微软等）财年始于7月，"fiscal Q1 2026" 可能是 2025年7-9月的旧数据。
  • 必须区分财年季度和日历季度。若无法确认，标注 "⚠️ 时间存疑"。

规则3：来源可信度分级（Source Credibility）

  • 🔴 禁止引用：Clarqo、Briefly、Quartz 等 AI 内容聚合/摘要站，如果其内容涉及尚未发生的事件（如财报结果发布于实际财报日之前），直接丢弃该条。
  • 🟡 谨慎引用：AI 生成的内容农场——如果同一事件在 BBC/Reuters/Bloomberg 等可信源中找不到交叉验证，直接丢弃。
  • 🟢 可信来源：Reuters、Bloomberg、BBC、WSJ、FT、CNBC、官方 IR 页面、公司新闻稿。

规则4：如实汇报（Honest Reporting）

  • 如果某条新闻无法确认时间或来源真伪，在报告中标注 "⚠️ 无法确认时间/来源"，而不是编造细节。
  • 宁缺毋滥：如果只有一个可靠来源，就只报一个。绝不为了凑篇幅而引用可疑来源。

规则5：交叉验证（Cross-Verification）

  • 涉及具体财务数据（营收、EPS、增速、涨跌幅）的新闻，必须有至少一个可信来源的交叉验证。
  • 如果所有来源都是 AI 内容农场或无法验证，该条丢弃。

翻译规则

  • 英文标题必须翻译为中文: 报告中出现的带 [EN] 标签的英文标题，必须翻译为中文，同时保留英文原标题在括号中。
  • 格式: 中文翻译（English Original Title）
  • 翻译由模型在生成报告时完成，news-processor.py 仅负责数据采集。

中英文新闻平衡规则

  • 每条报告必须至少包含3-5条 🇨🇳 中国公司/中国市场的新闻
  • 不要只关注美国科技巨头，中国企业（A股、港股、中概股）的动态必须占据显著篇幅
  • 如果某时段中国新闻偏少，可以适当多选取几条；如果重大中文新闻密集，可以超过英文数量

🔬 行业先行指标扫描（每报必查）

简报核心目标：帮 Phil 赚钱，不是了解国际形势。先行信号同时关注利好和利空——做空信号的交易价值与做多信号完全对等。

通用框架（适用所有行业）

信号类别通用问题可信数据源
产能/供给行业产能收缩/扩张？减产/转产/停产？公司财报/电话会、行业研究
价格/成本上游原材料/中间品价格异动？合约价趋势？行业定价基准
订单/需求龙头在手订单/backlog/交期变化？电话会transcript、供应链调研
capex/资本下游大客户资本开支调整？行业capex周期？大客户财报、行业capex指引

AI 全产业链监控

传导链路：上游硬件 → 中游云设施 → 下游应用（间隔2-4周）

🔴 上游硬件（最先感知）：

  • GPU出货/交期（NVDA）
  • HBM产能售罄年份（MU/SK Hynix/Samsung）
  • NAND/HDD库存水平（WDC/STX/SNDK）
  • 服务器CPU出货（INTC/AMD）
  • 电力/冷却基础设施

🟡 中游云设施（验证需求）：

  • 四大云厂商（AMZN/MSFT/GOOGL/META）capex指引
  • AI收入增速
  • 数据中心开工/电力合同

🟢 下游应用（验证商业化）：

  • 企业AI支出/订阅增速
  • AI终端设备出货
  • 自动驾驶/Robotaxi扩张
  • AI广告/搜索收入
  • 🇨🇳 中国AI应用：微信AI/百度AI/字节豆包用户量 → 小米AI手机/PC出货 → 比亚迪/小鹏智能驾驶

标注规则

  • 单一先行指标触发 → ⚡ [行业 先行信号]
  • 同一行业 2+ 指标同时触发 → 🔥🔥🔥 [行业 多重先行信号]
  • 跨行业共振（如芯片+云capex同时亮灯）→ 🚨🚨🚨 [全链条先行信号共振] — 最高级别预警

监控板块（非公司白名单）

公司列不完。以下为板块索引——实际监控从news-processor.py抓取的所有新闻动态匹配，不限所列名称。

AI上游硬件：GPU/加速器(NVDA/寒武纪/海光等)、HBM/DRAM(MU/SK Hynix/Samsung等)、NAND/HDD(WDC/STX等)、CPU(INTC/AMD/ARM等) 🇨🇳 中国半导体：AI芯片设计(寒武纪/海光等)、晶圆代工(中芯/华虹等)、服务器/算力(中科曙光/浪潮等)、存储(长江存储/长鑫等) AI中游云：超大规模(AMZN/MSFT/GOOGL/META)、🇨🇳 中国云(腾讯/阿里/百度/华为)、数据中心REITs AI下游：企业AI软件(PLTR/CRM等)、🇨🇳 AI应用(科大讯飞/金山等)、AI终端(苹果/小米)、自动驾驶(Waymo/PONY/百度/Tesla) 能源/电力：数据中心电力(VRT/CEG)、电网 消费/出海：电商(PDD/阿里/美团/京东)、新能源车(比亚迪/小米/蔚小理)

传导顺序：HBM内存（最先被抽空） → SSD/HDD存储 → CPU处理器（最后验证）。全链条传导时间约6-8周。

Report Output Format

以下为最终模板，每条报告严格按此格式输出。

顶部区域（必须）

📊 **资本市场绝对影响力简报 | YYYY-MM-DD HH:MM**

⚠️ **Delta说明**：vs 上一版（HH:MM）。X小时增量：[核心变化简述，5-8项关键词]

---

## 🔬 先行信号扫描

| 象限 | 板块 | 状态 | 信号 |
|------|------|------|------|
| 产能/供给 | ... | 🟢/⚡/🔥 | ... |
| 价格/成本 | ... | ... | ... |
| 订单/需求 | ... | ... | ... |
| capex/资本 | ... | ... | ... |

**综合判断**：[无异常 / ⚡单信号 / 🔥🔥🔥多重信号 / 🚨🚨🚨全链条共振]

---

主体区域：增量新闻 + 严谨推导

每条新闻按以下格式，逐条排列：

## 🟢/🔴/🟡 [增量类型 — 事件标题] emoji

**vs HH:MM版增量**：[一句话说明与上一版的差异]

**来源**：来源A（[阅读原文](真实URL)）/ 来源B（[阅读原文](真实URL)）🟢/🟡 可信度标注

**核心**：
- 关键数据/事实点1
- 关键数据/事实点2

**严谨推导**：
- 🟢/🔴/🟡 多角度推导
- 🇨🇳 如涉及中国市场，单独标注
- ⚡ 如涉及先行信号，单独标注

增量类型标签：

  • [增量更新] — 已在上版覆盖，本版有新的细节/数据/反应
  • [全新] — 上版未覆盖的新事件
  • [跟进] — 进展有限，做简要追踪

底部区域（必须）

## 📋 本版 vs HH:MM版 核心Delta总结

| 事件 | 方向 | HH:MM状态 | 当前状态 |
|------|------|-----------|----------|
| ... | 🟢/🔴 | ... | ... |

## 🔥 关键日历窗口

| 时间 | 事件 | 重要性 |
|------|------|--------|
| ... | ... | 🔥🔥🔥🔥🔥 |

---

⚠️ **防幻觉校验**：✅ 逐条确认...

⏰ 数据截止：...
*⚠️ 免责：基于公开信息，不构成投资建议。*