SuperAPI AI Directory
Partner中文AI newsFavorites
Skillsnews-pulse

news-pulse

General & other

View repository

Updated 10-16 01:35

About this skill

Codex adapter note

This skill is generated from skills/news-pulse.md so Claude Code and Codex users share one canonical workflow.

  • Treat $ARGUMENTS as the user's request in the current Codex thread.
  • When the source mentions Claude-only surfaces such as Task, Agent, WebSearch, Bash, Read, or Write, use the closest Codex capability available in this session: subagents when available, web search when needed, shell commands for local tools, and normal file edits for workspace files.
  • Use shared project tools from tools/ in this repository. Prefer running commands from the repository root with paths like python3 tools/financial_rigor.py ...; if the current thread starts outside the repo, locate the actual checkout path first instead of assuming a fixed home-directory path.
  • Before starting research, run the date command to confirm today's date; treat it as the baseline for "latest" data and state the data cutoff date in the report header. Never assume the current date from training data.
  • Preserve the research quality rules from AGENTS.md: cross-check financial data, use exact arithmetic tools for valuation/math, and clearly label uncertainty and source gaps.
Company News Pulse: Rapid attribution team for abnormal stock price moves

Conduct recent news reconnaissance and abnormal move attribution for $ARGUMENTS.This is not deep investment research, it is rapid intelligence response—the goal is to answer within 10 minutes: "What happened to this company recently? What is the real cause of the abnormal stock price move? Should the investment thesis be reviewed?"

Applicable scenarios

  • A held/watchlist stock rises/falls sharply (general trigger line: single day ±5%, one week ±10%)
  • The stock price moves abnormally after earnings, and you want to quickly figure out what the market is reacting to
  • You see a news headline, but are not sure whether it is noise or a real signal
  • Not applicable: full investment research (use /investment-team), in-depth earnings reading (use /earnings-review), long-term thesis tracking (use /thesis-tracker）

Execution process

Step 1: Confirm parameters and scenario

Clarify the following information with the user (if not provided in $ARGUMENTS):

Parameter Description Default
Company name Chinese/English/stock ticker all acceptable Required
Time window Number of days to look back when scouting news Default 14 days, can be shortened to 7 days during earnings season
Stock price abnormal move Rise/fall magnitude + time, e.g. "down 12%/3 days" Optional, if available use it to focus attribution
Focus area Company events / regulation / industry / sentiment Default equal weighting across the four areas

If the user only gives the company name, first ask back: "How many recent days of news? Is there a specific abnormal stock price move to explain?"—do not silently assume。

Step 2: Information availability grading

Refer to investment-team.md 's A/B/C rating, but the dimensions are different:

Level Characteristics Scouting strategy
A level (abundant information) Large-cap stocks, broad media coverage, earnings season The focus isnoise reduction and attribution—too much information actually makes it harder to find the real cause, each Agent must have judgment and filter out secondhand news that "repeats and relays"
B level (moderate information) Small and mid-cap, average coverage Standard mode, each key event should be accompanied by 1-2 independent sources
C level (scarce information) Hong Kong small caps, newly listed, obscure Switch to "beginner mode"—it may be impossible to find any news that can explain the abnormal move,this conclusion itself has value(it may be technical/liquidity rather than fundamental)

Tell each Agent the rating, influencing its scouting approach.

Step 3: Create the team

Use TeamCreate to create the team:

  • team_name: {公司名}-newspulse(English lowercase, e.g. pdd-newspulse）
  • agent_type: team-lead

Step 4: Create 4 scouting tasks

Use TaskCreate to create the following 4 tasks:

Task 1: Company event scouting (company-event-scout)

  • subject: 侦察 {公司名} 近 {N} 天的公司本体事件
  • description:
    1. Official Announcements: Recent disclosures on regulatory disclosure platforms such as HKEX/SEC/Juchao
    2. Financial Reports and Performance Guidance: Key points from the latest quarterly report/annual report, performance forecasts, and earnings calls
    3. Management Actions: Executive changes, increases/decreases in holdings, share buybacks, dividends, equity incentives
    4. Major Business Events: New product launches, mergers and acquisitions, business divestitures, major customers/large orders
    5. Capital Operations: Refinancing, convertible bonds, ADR conversion, proposals to return to A-shares/delist
    6. Litigation and Compliance: Being sued, self-disclosed compliance incidents
    7. Each event should be annotated with:Date / Source link / One-sentence summary / Possible relevance to stock price anomalies (High/Medium/Low)
    8. Output a timeline table in reverse chronological order

Task 2: Regulatory and Policy (regulatory-watcher)

  • subject: 侦察 {行业/公司} 近 {N} 天的监管与政策变化
  • description:
    1. Industry Regulation: New regulations, fines, rectification, license changes in the industry
    2. Cross-border Policy: China-US relations (China concept stocks), tariffs, export controls, data security
    3. Tax Policy: Changes related to value-added tax, corporate income tax, personal income tax
    4. Antitrust and Competition Law: Investigations, fines, merger vetoes
    5. Special Industry Policies: Pharmaceutical centralized procurement, education double reduction, real estate three red lines, internet platform regulation, etc.
    6. Monetary and Foreign Exchange: Exchange rate/interest rate/capital control changes affecting the company
    7. Each policy should be annotated with:Date / Source / Degree of direct impact on the company (Direct/Indirect/None)
    8. Key judgment: Whether a "policy black swan" has just landed

Task 3: Industry and Competitors (industry-peer-analyst)

  • subject: 侦察 {公司名} 行业格局与对手近 {N} 天的动态
  • description:
    1. Direct Competitors: List 3-5 core competitors, check recent events for each (financial reports, products, price wars, personnel)
    2. Industry Chain Upstream and Downstream: Upstream raw materials/suppliers, downstream customers/channels, recent changes in prices, capacity, orders
    3. Overall Industry: Industry prosperity data, shipments, demand-side signals (consumption data, bidding data, etc.)
    4. Threat of Substitutes: Impact of new technologies and new business models on the industry
    5. Industry Index Performance: Recent performance of stocks in the same industry, whether the company is outperforming/underperforming/in line
    6. Key judgment:Is this a company-specific event, or a beta fluctuation of the entire industry?
    7. Each event should be annotated with source and date

Task 4: Market Sentiment and Sell-side/Influential Voices (sentiment-tracker)

  • subject: 侦察 {公司名} 近 {N} 天的市场情绪与机构观点变化
  • description:
    1. Sell-side Rating Changes: Recent rating/target price adjustments by Goldman Sachs, Morgan, CICC, etc.
    2. Institutional Holdings Changes: 13F disclosures (US stocks), Hong Kong Stock Connect holdings, northbound capital flows
    3. Short Selling Data: Short interest ratio, newly released short-selling reports (if any)
    4. Influential Voices' Views: Can call python3 tools/xueqiu_scraper.py Capture recent related statements from Duan Yongping and other big Vs
      • Duan Yongping user_id: 1247347556
      • Command example:python3 tools/xueqiu_scraper.py --user-id 1247347556 --keywords {公司名},{股票代码} --output /tmp/dyp-{公司名}.md
      • Call only when the company is a target of interest to Duan Yongping/Li Lu; otherwise skip to save time
    5. Rumors and speculative posts: media-unconfirmed rumors, social media discussion hotspots (Xueqiu/X/Reddit)
    6. Technical signals: whether key support/resistance has been touched, whether there are block trades, abnormal margin trading
    7. Key judgment:Is it driven by fundamentals or by sentiment/capital flows?

Step 5: Launch 4 Agents in parallel

Must call the Task tool 4 times in parallel in the same message. Each Agent configuration:

  • subagent_type: general-purpose
  • run_in_background: true
  • team_name: {公司名}-newspulse
  • name: corresponding role name (company-event-scout / regulatory-watcher / industry-peer-analyst / sentiment-tracker)

Prompt template for each Agent:

你是 {公司名} 新闻脉搏团队中的"{角色中文名}"，负责侦察 {侦察维度} 维度的最近 {N} 天事件。

时间窗口：{起始日期} ~ {今日日期}
股价异动背景：{用户提供的异动信息，无则写"无特定异动，常规体检"}
信息可得性等级：{A/B/C}

请完成任务 #{任务编号}：{任务subject}

具体侦察要求：
{任务description的内容}

**侦察方法**：
- 优先使用 WebSearch 搜索时效性查询（关键词加日期或"最近"、"latest"、"2026"）
- 关键事件用 WebFetch 精读原始来源（公告原文、财报、监管文件）
- 对每个事件做"独立信源验证"——传言至少要 2 个独立来源
- **不要被标题党误导**：标题与正文不符的事件要标注"标题误导"

**输出格式（重要）**：
1. **核心发现**：3-5 条最关键的事件，每条 1-2 句话
2. **完整事件时间线表格**（按日期倒序）：
   | 日期 | 事件 | 来源 | 与股价异动相关性 | 持续性 |
3. **本维度归因结论**：基于侦察到的事件，回答"这个维度能否解释股价异动？置信度多少？"
4. **数据缺口声明**：哪些信息没找到、哪些有疑点、哪些需要等更多信息
5. 严格区分"事实"与"推测"，遵循 CLAUDE.md 客观性原则

**完成后**：
1. 使用 TaskUpdate 将任务标记为 completed
2. 通过 SendMessage 把完整侦察报告发送给 team-lead（type: "message", recipient: "team-lead"）

Step 6: Track progress in real time

  • Each time a reconnaissance report is received, show the user the 3 core findings for that dimension
  • Wait for all 4 reports to arrive
  • After all have arrived, send shutdown_request to the 4 Agents via SendMessage

Step 7: team-lead comprehensive attribution

Summarize the 4 reconnaissance reports and outputAbnormal Move Attribution Report(not a research report; the focus is "judgment"):

1. One-sentence attribution

Use one paragraph (30-60 characters) to explain: the main cause of this stock price abnormal move + secondary cause + nature judgment (value event/sentiment fluctuation/unclear)

2. Complete event timeline (merging 4 dimensions)

In reverse chronological order, merge events from all dimensions:

Date Dimension Event Source Abnormal Move Attribution Weight
2026-04-30 Company XX Link 🔴 High
2026-04-29 Industry XX Link 🟡 Medium
2026-04-28 Sentiment XX Link ⚪ Low

Weight legend: 🔴 High (sufficient to explain the abnormal move on its own) / 🟡 Medium (contributes part of it) / ⚪ Low (background noise)

3. Abnormal Move Attribution Table

Candidate explanation Evidence Counter-evidence Confidence Persistence
Example: earnings miss Revenue 5% below expectations, gross margin decline One-off factors explained by management High Short term 1-2 weeks
Example: industry beta Peers fell 8% over the same period This stock's decline is clearly larger than the industry's Medium In sync with the industry

4. Nature judgment (core conclusion)

Check one:

  • Value event: a real change has occurred in fundamentals (performance, moat, management, endgame), requiring a review of the investment thesis
  • Sentiment/technical fluctuation: no change in fundamentals; driven by capital flows/sentiment/Beta, can be viewed as an opportunity or noise
  • True cause unclear: no event can be found that matches the magnitude of the stock price abnormal move—This is the most dangerous conclusion, either the market knew something in advance (insider trading/front-running), or we missed an information source
  • Mixed: partly value events + partly sentiment amplification

5. Reconnaissance summary by dimension

3-5 most important findings per dimension + attribution contribution of that dimension.

6. Action recommendations

Action Recommended or not Reason
Trigger investment thesis review (/thesis-tracker）
Trigger in-depth financial report study (/earnings-review）
Trigger management review (/management-deep-dive）
Position adjustment action (add/reduce/hold) For reference only; final decision-making authority rests with the user
Observe only

7. Tracking checklist for the next 7-30 days

  • Pending disclosure event 1 (e.g., 5/15 earnings call)
  • Indicator to track 2
  • Key observation signal 3

8. Information gap statement

Honestly list the doubts that this reconnaissance could not resolve, the information that could not be found, and the matters that require more disclosure.Rather mark it as "uncertain" than fill it in with speculation。

Step 8: Save the report

Write to reports/{公司名}/{公司名}-news-{YYYYMMDD}.md. If reports/{公司名}/ directory does not exist, create it (indicating that no research report has ever been created for this company).

Step 9: Clean up the team

Use TeamDelete to clean up team resources.

Key principles

  1. Fast beats complete—the core value of this skill is to give an attribution judgment within 10-15 minutes; do not get stuck in deep analysis (that is the responsibility of other skills)
  2. Attribution takes priority over listing—finding events is not difficult; what is difficult is judging "which event deserves to be the cause of this stock price anomaly." Do subtraction, not addition
  3. Face "unknown" honestly—when the main cause cannot be found, clearly write "the true cause is unknown." This is more valuable than forcing a causal chain (the market may be front-running negative news)
  4. Do not presuppose a position—do not lean toward "this is emotional fluctuation, nothing serious" just because you hold a position. Write whichever side the evidence points to
  5. Distinguish "catalyst" from "coincidence"—events happening at the same time are not necessarily the main cause of the anomaly; check whether the magnitude of impact matches
  6. Respect information availability—a Tier C company may simply have no news that can be found, and this conclusion itself must be written out
  7. Follow the CLAUDE.md objectivity principle—attach data sources to all judgments, distinguish facts from opinions
  8. Do not make decisions for the user—provide a list of attribution and action recommendations, but buy and sell decisions are made by the user