P News Briefing

Depth

• deep (default): News + market/people reactions, sourced links, quotes, ~5 min read
• quick: Headlines + brief context, 5-10 bullets with links, ~2 min read

Flow

1. Parse input: Extract topic(s) and depth. Default depth is deep.
2. Normalize topics: Convert to URL-safe slugs for filenames (e.g., "artificial intelligence" -> ai, "open source" -> open-source).
3. Create date directory: {notebook_daily_dir}YYYY-MM-DD/news/
4. Deduplicate against previous briefings (parallelize if possible): Before researching, check for existing briefings on the same topic-slug in recent daily/*/news/ directories (scan the last 14 days). Run dedup scans for all topics in parallel. Read any found files. Stories already covered in a previous briefing should be skipped entirely unless there is a meaningful follow-up (e.g., new data, reversal, sequel event). When a follow-up exists, write it as its own section and briefly note it builds on prior coverage — do not repeat the original story.
5. Research directly using WebSearch. Process all topics in parallel (parallelize if possible): after the dedup scan completes for all topics, dispatch all per-topic research tasks concurrently as a fan-out, then collect results and summarize.

Per-Topic Research (parallelize if possible)

IMPORTANT: Recency is critical. Follow these rules strictly:

• Always use today's exact date (YYYY-MM-DD) to construct queries. Never use just a year or month alone.
• Prefer "today", "yesterday", "past 24 hours", "past 48 hours" in queries — these terms signal freshness to search engines far better than month/year.
• Discard stale results: After gathering search results, check publication dates. Only include stories from the last 7 days (for deep) or last 3 days (for quick). If a result has no visible date, deprioritize it.
• If initial results are stale, run follow-up queries with stricter time language (e.g., "today", "yesterday", site:reuters.com OR site:bloomberg.com).

For deep depth:

Phase 1 - Breaking/recent news: Run 3 WebSearch queries in parallel:

• "[TOPIC] news today [Full Month Day, Year]" (e.g., "AI news today February 6, 2026")
• "[TOPIC] latest news this week [Month Year]"
• "[TOPIC] breaking developments [Month Day Year]"

Phase 2 - Reactions & analysis: Run 2-3 WebSearch queries in parallel:

• "[TOPIC] market reaction today [Month Year]"
• "[TOPIC] social reaction today [Month Year]"
• "[TOPIC] analyst reaction [Month Day Year]"
• "[TOPIC] expert opinion latest [Month Year]"

Phase 3 - Fill gaps (only if needed): If Phase 1-2 returned fewer than 3 distinct stories, run 1-2 more targeted queries:

• "[TOPIC] [specific subtopic from earlier results] [Month Day Year]"
• site:reuters.com OR site:apnews.com "[TOPIC] [Month Year]"

Graceful failure: If a WebSearch query fails or returns no results, note the gap and continue with available data. Include a footer note in the output listing any failed queries so the reader knows coverage may be incomplete.

Phase 4 - Write output: Combine into structured markdown with:

• YAML frontmatter: topic, date, depth
• Major stories as H2 sections (each must include publication date)
• Market data (stock moves, valuations) where relevant
• Named quotes from analysts, CEOs, social media
• Inline source links as markdown hyperlinks
• A "Big Picture" H2 summary at the bottom

For quick depth:

Run 3 WebSearch queries in parallel:

• "[TOPIC] news today [Full Month Day, Year]"
• "[TOPIC] headlines this week [Month Year]"
• "[TOPIC] latest [Month Day Year]"

Then write a concise markdown file with:

• YAML frontmatter: topic, date, depth
• 5-10 bullet points, each a headline with 1-2 sentences of context
• Each bullet must include the publication date (e.g., "Feb 5")
• Inline source links as markdown hyperlinks
• Discard any results older than 3 days

Output Format

---