tech-market-ai-radar

name	tech-market-ai-radar
description	Collect, verify, deduplicate, rank, and summarize current hot news and intelligence across technology, AI, public markets, listed companies, macro policy, chips, software, startups, and internet platforms. Use when Codex is asked to gather "全网热点", daily/weekly tech news, stock-market moving news, AI industry updates, investment-watch briefs, topic plans, or source-linked news digests.

Tech Market AI Radar

Overview

Use this skill to produce source-linked intelligence briefs about technology, AI, and markets. Treat "全网" as broad, multi-source coverage rather than literal exhaustive crawling.

For current events, always verify with live sources. Prefer primary sources, official disclosures, regulators, exchanges, company blogs, model/research announcements, earnings materials, and reputable financial or technology media.

Default Scope

If the user does not specify scope, use:

• Time window: past 24 hours.
• Language: Chinese output.
• Geography: global, with emphasis on China and the United States.
• Tracks: AI, big tech, chips, software/SaaS, consumer internet, listed-company news, stock movers, macro/regulation, funding/M&A.
• Output: concise intelligence brief with links, dates, credibility labels, and market/industry impact.

Ask a clarifying question only when the missing choice changes the result materially, such as a requested stock market, a paid/private source requirement, or a publication-specific format.

Collection Workflow

1. Define the time window and user objective.
2. Build a query matrix across tracks, regions, and source types.
3. Search broadly, then open and inspect the most relevant primary or reputable sources.
4. For market-moving claims, verify with at least one primary source or two independent reputable sources when possible.
5. Normalize each item into the item schema below.
6. Deduplicate items that describe the same event.
7. Rank by impact, credibility, recency, novelty, and user relevance.
8. Produce a brief that separates facts, interpretation, and open questions.

Use references/source-playbook.md for source classes, query patterns, credibility tiers, and finance-specific handling. Use references/output-formats.md for reusable brief formats.

Item Schema

When collecting items, normalize them as JSON objects before ranking:

{
  "title": "short event title",
  "summary": "one or two sentence factual summary",
  "url": "https://source.example/article",
  "source": "source name",
  "published_at": "2026-06-03T09:30:00+08:00",
  "category": "AI | stocks | chips | software | policy | macro | funding | product",
  "region": "US | CN | EU | global",
  "tickers": ["NVDA", "MSFT"],
  "credibility": "primary | high | medium | low",
  "impact": "high | medium | low",
  "novelty": "high | medium | low",
  "attention": "high | medium | low",
  "event_key": "optional stable dedupe key"
}

Do not invent dates, prices, stock moves, funding amounts, product details, or source links. If a field is unknown, omit it or mark it as unknown in prose.

Ranking Rules

Prioritize items that are:

• Primary-source verified or covered by reputable financial/technology media.
• Likely to affect listed companies, sectors, regulation, product adoption, or AI capability curves.
• New inside the requested time window.
• Connected to companies, tickers, models, chip supply chains, cloud platforms, or policy regimes the user cares about.

Down-rank:

• Unsourced social-media rumors.
• Reposted articles with no new facts.
• Old stories resurfacing without a new development.
• Promotional content without independent confirmation.

Scripted Ranking

Use scripts/rank_news_items.py when you have a JSON array of normalized items and want deterministic deduplication and ordering:

python scripts/rank_news_items.py items.json --format markdown --limit 15

The script accepts a JSON array or JSONL, computes a score, deduplicates by event_key or normalized title, and outputs either Markdown or JSON.

Briefing Standards

Every final brief should include:

• A date/time range in absolute terms.
• Source links for each major item.
• A clear distinction between fact and analysis.
• A brief impact line for important stories.
• A watchlist of uncertain but important developments when relevant.

For finance or stock topics, include a short caveat that the brief is informational and not investment advice. Use current price/market data tools when the user asks about stock moves, prices, market cap, earnings reaction, or trading implications.

Quality Checklist

Before finalizing:

• Confirm the time window and dates are consistent.
• Check that each top item has a working source link.
• Remove duplicates and near-duplicates.
• Avoid treating commentary, predictions, or rumors as facts.
• Label confidence when evidence is incomplete.
• Keep the brief scannable; the user should see the top stories and why they matter within the first screen.