# How the News-Pulse Skill Attributes Stock Price Movements in AI-Berkshire > Unlock stock price movement insights with the News-Pulse skill in AI-Berkshire. Discover how four scouting agents analyze news and sentiment for actionable recommendations. - Repository: [Xbt Lin/ai-berkshire](https://github.com/xbtlin/ai-berkshire) - Tags: deep-dive - Published: 2026-06-27 --- **The News-Pulse skill attributes stock price movements by deploying four parallel scouting agents—company-event, regulatory, industry-peer, and sentiment trackers—to gather evidence from multiple dimensions, then synthesizing findings into a confidence-weighted attribution report with actionable recommendations.** The `ai-berkshire` repository implements a sophisticated multi-agent system for financial analysis. The News-Pulse skill specifically addresses the challenge of explaining why a stock moved by orchestrating parallel research tasks across distinct information domains, ensuring every attribution claim meets rigorous evidence standards defined in [`skills/news-pulse.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/news-pulse.md). ## The Four-Agent Attribution Framework At the core of the News-Pulse skill are four specialized agents, each targeting a specific attribution dimension to isolate causal factors behind price swings. ### Company Event Scout The `company-event-scout` agent investigates official corporate disclosures, including earnings releases, management actions, mergers and acquisitions, and litigation filings. This agent focuses on **value events**—fundamental changes in the company's business trajectory that directly impact intrinsic valuation rather than temporary market sentiment. ### Regulatory Watcher The `regulatory-watcher` agent monitors policy shifts that could affect the stock's operating environment. It tracks new regulations, tax code changes, antitrust actions, and cross-border policy adjustments that create systematic risk or opportunity for the specific equity. ### Industry Peer Analyst The `industry-peer-analyst` agent examines sector-wide dynamics and competitor movements. It scouts supply-chain disruptions, peer earnings surprises, sector beta-wave signals, and industry-specific data releases that drive relative performance against benchmarks. ### Sentiment Tracker The `sentiment-tracker` agent captures market psychology through sell-side rating changes, institutional holding shifts, short-interest fluctuations, and influential commentary from market participants and social media sources. ## The 8-Step Attribution Workflow The News-Pulse skill executes a systematic workflow orchestrated through the specification in [`skills/news-pulse.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/news-pulse.md): 1. **Parameter Validation** – The skill validates the company name, time window (defaulting to 14 days), and any explicit price-move description provided by the user. 2. **Information Availability Grading** – The system assigns an A/B/C grade (A for well-covered large caps, C for information-scarce stocks) to inform each agent about source material richness. 3. **Team Creation** – A temporary team named `company-newspulse` is instantiated with a `team-lead` agent responsible for aggregation. 4. **Parallel Task Dispatch** – Four `TaskCreate` calls launch the scouting agents simultaneously, with each receiving dimension-specific instructions. 5. **Agent Investigation** – Each agent executes a three-phase research protocol: **WebSearch** for recent time-bounded news, **WebFetch** for source verification, and cross-reference validation requiring at least two independent sources for any claim. 6. **Structured Output Generation** – Agents return timeline tables containing date, event, source URL, and relevance ratings (high/medium/low). 7. **Real-Time Aggregation** – As reports arrive, the team lead surfaces the three most salient findings per dimension while waiting for completion. 8. **Final Synthesis** – The lead compiles a unified attribution report including a one-sentence core cause, secondary factors, merged timelines, and a confidence-weighted evidence table. ## Evidence Standards and Confidence Grading The News-Pulse skill enforces strict evidentiary standards. Each agent must verify claims through **at least two independent sources** before inclusion. The final synthesis produces an **attribution table** weighing evidence against counter-evidence, resulting in a confidence assessment that clearly flags when causation remains uncertain (nature of move: value event vs. sentiment vs. unknown). ## Persistence and Cleanup Upon completion, the final markdown report is persisted to `reports//-news-.md` following the pattern defined in the skill specification. The temporary `company-newspulse` team is then deleted to maintain system hygiene. ## How to Invoke the News-Pulse Skill Users interact with the News-Pulse skill through slash commands defined in [`codex-prompts/news-pulse.md`](https://github.com/xbtlin/ai-berkshire/blob/main/codex-prompts/news-pulse.md): ```markdown # Simple invocation with default 14-day window /news-pulse Tencent ``` ```markdown # Explicit price-move context (down 12% over 3 days) /news-pulse PDD down 12% 3 days ``` The skill interprets natural language price movements and time windows, converting them into structured parameters for the agent workflow. ## Output Structure and Deliverables The generated attribution report (`reports//-news-.md`) contains: - **Core Attribution**: A one-sentence explanation of the primary driver (value event, sentiment, or unknown) - **Unified Timeline**: Merged chronology of all dimension-specific events - **Confidence Weighting**: Evidence versus counter-evidence assessment - **Actionable Recommendations**: Specific next steps such as triggering `/thesis-tracker` reviews or initiating deeper earnings analysis ## Summary - The News-Pulse skill uses **four parallel agents** to investigate stock price movements across company events, regulatory changes, industry dynamics, and market sentiment. - **Strict evidence requirements** mandate at least two independent sources per claim, with A/B/C grading indicating information availability. - The workflow produces **confidence-weighted attribution reports** saved to `reports//-news-.md` with actionable recommendations. - Users invoke the skill via `/news-pulse` commands with natural language price-move descriptions. ## Frequently Asked Questions ### What makes the News-Pulse skill different from simple news aggregation? The News-Pulse skill employs **parallel multi-agent scouting** with specific attribution dimensions rather than keyword-based aggregation. Each agent specializes in distinct causal factors (regulatory, industry, sentiment, corporate events), and the system requires **verified dual-source confirmation** before accepting any explanatory factor, producing evidence-grade attribution rather than narrative summaries. ### How does the skill handle information-scarce stocks? For stocks with limited coverage, the skill assigns a **Grade C information-availability rating** during the validation phase. This signals to all four agents that source material may be sparse, adjusting research protocols while maintaining the same evidentiary standards. When sufficient evidence cannot be obtained, the final report explicitly flags the cause as **unknown** rather than inferring unsubstantiated explanations. ### What file contains the canonical specification for the News-Pulse workflow? The complete workflow specification, including agent definitions, prompt templates, and output formats, resides in **[`skills/news-pulse.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/news-pulse.md)** within the `xbtlin/ai-berkshire` repository. The [`codex-prompts/news-pulse.md`](https://github.com/xbtlin/ai-berkshire/blob/main/codex-prompts/news-pulse.md) file provides the user-facing slash command interface that forwards requests to this canonical specification. ### Can the News-Pulse skill trigger other analysis workflows? Yes. The final synthesis phase generates **actionable recommendations** that can automatically trigger downstream skills. For example, the skill may recommend triggering `/thesis-tracker` for portfolio review or initiating deeper earnings analysis when the attribution reveals fundamental value events requiring further investigation.