news-sentiment

News Sentiment Analysis

Purpose

Fetch financial news articles for a company published in the past 24 hours and score each headline for sentiment. Produce a news channel score from -10 (very bearish) to +10 (very bullish).

Step 1 — Resolve Ticker and Company Name

If a ticker is provided, confirm the company name via WebSearch: {ticker} stock company name site:finance.yahoo.com OR site:stockanalysis.com

If only a company name is provided, resolve the ticker the same way.

Step 2 — Read Config for NewsAPI Key

Run:

cat ${CLAUDE_PLUGIN_ROOT}/config.json 2>/dev/null

If the file exists and contains newsapi.key (i.e., config["newsapi"]["key"]), proceed to Step 3a. Otherwise proceed to Step 3b (WebSearch fallback).

Step 3a — Fetch via NewsAPI (key available)

WebFetch the following URL, substituting:

• {ticker}: the stock ticker symbol
• {company_name}: URL-encoded company name
• {ISO_24h_ago}: current UTC time minus 24 hours in format YYYY-MM-DDTHH:MM:SSZ
• {KEY}: the resolved string value of config["newsapi"]["key"]

https://newsapi.org/v2/everything?q={ticker}%20OR%20%22{company_name}%22&from={ISO_24h_ago}&sortBy=publishedAt&language=en&pageSize=20&apiKey={KEY}

Extract from each article: title, description, source.name, publishedAt, url. Use the first 20 results.

Step 3b — WebSearch Fallback (no key)

Run WebSearch: "{company_name}" OR "{ticker}" stock news site:reuters.com OR site:bloomberg.com OR site:seekingalpha.com OR site:finance.yahoo.com

Extract up to 20 headline + description pairs from search results. Record source as "WebSearch" and note in output: "News via WebSearch — no NewsAPI key configured"

Step 4 — Score Headlines

For each article headline + description pair, assign a score from -1.0 to +1.0:

• +0.8 to +1.0: Record earnings beat, major upgrade, regulatory approval, significant partnership
• +0.4 to +0.7: Positive guidance, price target raise, solid quarterly results, market share gains
• 0.0: Neutral announcements, routine filings, personnel changes with no clear valuation impact
• -0.4 to -0.7: Missed estimates, minor regulatory concern, analyst downgrade
• -0.8 to -1.0: Major investigation, fraud allegation, severe earnings miss, CEO departure under fire

Score the combination of headline + description together. A positive headline with a cautionary description should score lower than the headline alone suggests.

Step 5 — Compute Channel Score

news_score = mean(all_headline_scores) × 10

Identify the 3 articles with the highest absolute score value (strongest signal, regardless of direction — these are the most signal-rich stories). These are the "key headlines" to surface in output.

Step 6 — Handle Low Volume

If fewer than 5 articles were found, add this warning to output: "⚠️ Low news volume ({n} articles) — score may not be representative"

Step 7 — Output

Return the following structured result (consumed by sentiment-report and sentiment-analyst):

NEWS SENTIMENT RESULT
Ticker: {ticker}
Company: {company_name}
Channel Score: {news_score} / 10
Signal: {label per signal bands below}
Article Count: {n}
Source: {NewsAPI | WebSearch}
Low Volume Warning: {yes | no}

Key Headlines (highest absolute sentiment score):
1. [{score}] "{title}" — {source_name}
   {url}
2. [{score}] "{title}" — {source_name}
   {url}
3. [{score}] "{title}" — {source_name}
   {url}

Signal bands:

• score >= +7.0 → Very Bullish
• score >= +3.0 and < +7.0 → Bullish
• score > -3.0 and < +3.0 → Neutral
• score <= -3.0 and > -7.0 → Bearish
• score <= -7.0 → Very Bearish