Web Research Brief

Structured web research and synthesis. Use when the user needs to investigate a topic, compare options, gather sources, or produce a research brief. Triggers on phrases like "research", "look into", "find out about", "compare", "what's the latest on", "deep dive", or "literature review".

Research Assistant

Systematic web research → source extraction → structured synthesis.

Workflow

1. Scope — Clarify the research question and constraints (time range, region, language, depth). If ambiguous, infer from context and state assumptions.
2. Search — Use web_search with targeted queries. Run 2-4 queries with varied phrasing to reduce blind spots. Prefer freshness filter for time-sensitive topics.
3. Fetch — Use web_fetch on the most relevant results. Extract key claims, data points, and source URLs. Skip paywalled or low-value pages quickly.
4. Synthesize — Organize findings into a structured brief (see Output Format).
5. Cite — Every factual claim must include a source URL. Flag unsourced claims explicitly.

Search Strategy

• Start broad, then narrow with domain-specific terms
• For comparisons: search each option independently, then "X vs Y" queries
• For technical topics: include version numbers or date ranges
• For regional info: set country and language parameters on web_search
• If initial results are thin, rephrase or switch to a different angle before concluding "no results"

Output Format

# Research Brief: [Topic]

## Summary
2-4 sentence executive summary.

## Key Findings
1. **[Finding]** — [1-line detail]. [Source](url)
2. **[Finding]** — [1-line detail]. [Source](url)

## Comparison (if applicable)
| Criterion | Option A | Option B |
|-----------|----------|----------|
| ... | ... | ... |

## Open Questions
- [Unresolved item 1]
- [Unresolved item 2]

## Sources
- [Title](url) — retrieved [date]
- [Title](url) — retrieved [date]

Quality Checks

• At least 3 distinct sources for any claim
• No fabricated URLs — every link must come from actual search/fetch results
• Flag recency: note if a source is > 1 year old
• If results are contradictory, present both sides rather than picking one