research-deep-dive

name	research-deep-dive
description	Multi-source research on any topic — run targeted searches, fetch sources, evaluate quality, compile findings into a structured markdown report file. Use for "research X", "write a report on Y", "deep dive into Z".
category	research
version	1.0.0
examples	[{"prompt":"Deep dive into the market for personal AI assistants with evidence links."}]
test_commands	["python -m pytest tests/test_skills.py"]

Research Deep Dive

Procedure

1. Plan searches — Break the topic into 2-4 key questions or angles. Run focused web_search calls for each. Don't repeat the same query — vary the wording to get different sources.

2. Fetch depth — For the most promising results, use fetch_url to get full page content where the auto-fetch snippet was insufficient.

3. Synthesize — Compare sources, note contradictions, identify authoritative sources, flag uncertainty.

   Build a claim-level evidence map before writing:

   • Claim
   • Supporting source URLs
   • Source freshness and authority
   • Contradictions or caveats
   • Confidence: high, medium, or low

4. Write report — Use write_file to save a structured markdown report covering:

   • Summary (key findings in 2-3 paragraphs)

   • Key Findings (detailed breakdown per research angle)

   • Sources (numbered list with URLs and why each is credible)

   • Evidence Map (claims linked to source numbers)

   • Open Questions (what's unclear or contradictory)

5. Report back — Tell the user where the report was saved and the top 3 findings.

Rules

• At least 2 web_search calls before writing — one search is not deep.
• Always include source URLs in the report.
• Flag uncertainty explicitly: "Source X says A, but source Y says B — X is more recent (2026) so that's more reliable."
• Do not use your own knowledge instead of search for current topics.