Auto Research

Auto-Research

Description

Deep research on any topic using multiple sources, with citations and explicit gaps.

Trigger

Questions about topics, companies, people, markets, trends, technologies, or any "what do we know about X" question.

Process

1. Scope the question — What specifically does the user need? A quick answer, a briefing, or a deep dive? Default to briefing (5-10 key points).
2. Check internal memory first:
   • secondBrain.search for what we already know
   • Surface any prior research or decisions on this topic
3. Web research:
   • web_search for current facts, news, data
   • Run 2-3 queries with different angles (e.g., "[company] revenue", "[company] competitors", "[company] criticism")
4. Social signal check:
   • x_read for recent public sentiment, announcements, hot takes
   • Look for what practitioners say, not just press releases
5. Synthesize with citations:
   • Lead with the answer, not the process
   • Inline citations: "[Source: URL]" for every factual claim
   • Separate facts from analysis
6. Identify gaps — What couldn't be found? What's uncertain? What would require deeper digging?
7. Store findings:
   • secondBrain.search key facts for future retrieval
   • Offer to save a research brief to vault if the topic is important

Output

• TL;DR (2-3 sentences)
• Key findings (5-10 bullets, cited)
• Gaps and uncertainties (2-3 bullets)
• Suggested follow-ups (1-2 questions to dig deeper)

Examples

1. "What's happening with AI agents in healthcare?" — Memory check (prior interest in Project Beta), web search for market size + key players + regulation, X check for practitioner sentiment. Output: market briefing with 8 findings, 2 gaps (pricing data unclear, regulation still evolving).
2. "Tell me about this person before my meeting" — Memory check for past interactions, web search for LinkedIn/bio/company, X for recent posts. Output: 1-page briefing with role, company, recent activity, talking points.

Failure Modes

• Stale data — Web results may be months old. Always note dates on findings. Flag if freshness matters.
• Source quality — Prefer primary sources (company blogs, SEC filings, official docs) over aggregator articles. Flag when sources are thin.
• Scope creep — User asks about "AI" broadly. Narrow it: "Which angle? Market size, technical trends, investment, or specific companies?"
• API failures — If web_search or x_read fails, say so explicitly. Don't hallucinate findings. Fall back to secondBrain.search and prior knowledge with clear caveats.