gaos6e/ai-daily-brief

AI Daily Brief

Purpose

Generate a Chinese AI daily brief that stays fresh, source-traceable, and merged by topic. Keep official sources first. Use authoritative media only when official coverage is sparse or when media materially advances the story.

Load Order

• Read references/sources.md before collecting headlines. That file is the source registry.
• Read references/quality-rules.md before filtering, ranking, deduplicating, or writing. That file contains the editorial rules and output contract.
• Run scripts/check_sources.py when source health is unknown, after adding a source, or before relying on a new recurring job.

Default Contract

• Use the last 24 hours as the primary coverage window, anchored to run time.
• Output Chinese only.
• Write one paragraph per merged topic.
• Do not output raw links unless the user explicitly asks.
• Target 8-12 merged items on normal days. Allow 4-20 depending on actual activity.
• Include a header with run date and coverage window.
• Keep fewer items rather than padding with weak material.

Workflow

1. Set the primary coverage window to the last 24 hours.
2. Collect tier-1 official machine-readable sources first. Prefer RSS/feed URLs from references/sources.md.
3. Collect tier-2 official page sources next. Use browser for JS-heavy pages and web_fetch for static pages.
4. Supplement with tier-3 authoritative media only when official coverage produces fewer than 4 merged topics, or when an official story needs corroboration and market context.
5. When a page-only official source is hard to scrape, use the official-domain search patterns from references/sources.md rather than drifting to generic search.
6. Record candidate items with source, title, timestamp, url, source_type, and a one-line raw note before summarizing.
7. Filter candidates using references/quality-rules.md. Do not auto-keep weak corporate marketing just because it came from an official domain.
8. Merge same-topic items before ranking. Prefer the official framing when an official source and media cover the same event.
9. Rank merged topics using the impact rubric from references/quality-rules.md.
10. Write the final brief using the output contract. State when the brief had to rely on media supplementation or a widened secondary check.

Scarcity and Failure Policy

• If official sources yield fewer than 4 merged topics in 24 hours, supplement with authoritative media in the same 24-hour window.
• If fewer than 4 merged topics remain even after media supplementation, allow a secondary 72-hour official-only lookback for context, and label those items as 近72小时官方补充.
• If two or more tier-1 official sources are unavailable, continue with the remaining official sources plus authoritative media, but do not fabricate missing coverage.
• If a source conflicts with an official statement, prefer the official statement for factual claims and use media only for market context.
• If coverage is thin, say so. Do not fill quota with weakly related items.

Topic-Merge Rules

• Treat items as the same story when the entity, core action, and object match inside the coverage window.
• Merge cross-report variants such as OpenAI desktop superapp, desktop app combining chat and coding, and OpenAI superapp plan into one topic.
• Keep one canonical paragraph per topic.
• Combine sources as 来源：OpenAI / Bloomberg / WSJ.
• When only media cover the story, prefer the clearest outlet framing and note multi-outlet confirmation when available.

Writing Rules

• Start each paragraph with the event, not the source list.
• Give the new fact first, then the implication.
• Keep each paragraph to 1-3 sentences.
• Avoid title-plus-bullet formatting.
• Avoid raw URLs.
• Avoid filler like 值得关注 unless the implication is stated concretely.

Output Skeleton

Use a header like:

AI 大事日报 | 2026-03-20 10:30
统计窗口：2026-03-19 10:30 至 2026-03-20 10:30
说明：官方优先；官方更新偏少时才补充 Reuters / Bloomberg / WSJ / FT / The Verge / TechCrunch / The Information。

Then write one paragraph per merged topic. Keep the source list inside the paragraph tail, not as a separate bullet.

Additional Resources

• references/sources.md - Exact source registry, tiers, fallback queries, and verification notes.
• references/quality-rules.md - Inclusion rules, merge rules, ranking rubric, scarcity policy, and output contract.
• scripts/check_sources.py - Fast source-health check for the machine-readable and page-only sources in this skill.