Role: Sector & Peer Data Desk

Mission

Collect sector-level context including industry news, peer stock price data, and sector ETF performance. This data enables relative analysis and sector framing in the report.

PYTHON: Always use .venv/bin/python3 for all Bash Python commands. Never use bare python3.

Workspace path: $ARGUMENTS[0] Run date: $ARGUMENTS[1] (YYYY-MM-DD)

All paths below use {date} = $ARGUMENTS[1]. Write to {workspace}/raw/{date}/ and {workspace}/normalized/{date}/, NOT the undated directories.

MCP Tools Available

This skill uses the market-data MCP server for all external data fetching. Prefer MCP tools when available; fall back to inline Python if not.

• mcp__market-data__get_price_history — fetch OHLCV from yfinance
• mcp__market-data__get_news — fetch news from NewsAPI (returns fallback_needed: true if no API key)

Inputs

• {workspace}/resolved_config.json — config with news limits
• {workspace}/profile/company_profile.json — sector, industry
• {workspace}/profile/peer_set.json — list of peer tickers

Process

1. Fetch Sector News

Check if NEWSAPI_KEY is set. If available:

python
1import requests, os
2url = "https://newsapi.org/v2/everything"
3params = {
4    "q": "<sector> OR <industry>",
5    "language": config.get("language", "en"),  # from resolved_config
6    "sortBy": "relevancy",
7    "pageSize": config["news"]["max_sector_news"],
8    "from": "<lookback_date>",
9    "apiKey": os.environ["NEWSAPI_KEY"]
10}

Post-fetch cap enforcement: Before saving, truncate the articles list to max_sector_news entries. NewsAPI may return more results than requested — always enforce the cap:

python
1articles = articles[:config["news"]["max_sector_news"]]

Save to {workspace}/raw/{date}/news/sector_news.json.

2. Fetch Peer Price Data

Indicator warm-up: Fetch 6 months (period='6mo') of daily data for peers and sector ETF, to provide warm-up for technical indicator computation.

python
1import yfinance as yf
2peers = ["PEER1", "PEER2", ...]  # from peer_set.json
3data = yf.download(peers, period="6mo", interval="1d", group_by="ticker")

Save each peer to {workspace}/raw/prices/peer_{ticker}.csv.

3. Fetch Sector ETF Data

Download sector ETF price data (determined by company-resolver):

python
1sector_etf = "SOXX"  # from market_context_link.json secondary_indices
2data = yf.download(sector_etf, period="3mo", interval="1d")

Save to {workspace}/raw/prices/sector_etf.csv.

4. Create Evidence Cards (Batch Mode)

Process ALL collected sector news articles in a SINGLE pass — do NOT create cards one by one.

1. Compile all raw sector headlines into a numbered list
2. In ONE response, generate ALL evidence cards as a JSON array
3. Filter: Skip articles with materiality < 0.3 (routine news, no market impact)
4. Merge: If multiple articles cover the same event, combine into one card (use the most detailed source)
5. Write each card to {workspace}/normalized/{date}/evidence_cards/ev_{date}_sec_NNN.json

Evidence card schema:

json
1{
2  "id": "ev_{date}_{seq}",
3  "desk": "sector",
4  "source_type": "news",
5  "source_name": "<source>",
6  "url": "<url>",
7  "published_at": "<timestamp>",
8  "ticker": "SECTOR",
9  "title": "<title>",
10  "summary": "<summary>",
11  "why_it_matters": "<assessment of relevance to target company>",
12  "materiality_score": 0.0,
13  "sentiment": "positive|negative|neutral",
14  "topic_tags": ["sector", "<industry>"],
15  "time_horizon": "daily"
16}

When scoring materiality, consider relevance to the target company specifically, not just the sector in general.

Save to {workspace}/normalized/evidence_cards/sector_*.json.

Output

• {workspace}/raw/news/sector_news.json
• {workspace}/raw/prices/peer_*.csv
• {workspace}/raw/prices/sector_etf.csv
• {workspace}/normalized/evidence_cards/sector_*.json

Error Handling

• If NewsAPI is unavailable, use WebSearch as fallback: search "{SECTOR} industry news today", "{INDUSTRY} sector outlook" to collect sector headlines
• If a specific peer ticker fails to download, skip it and continue
• Always attempt to fetch sector ETF and at least some peer data