Research

Research & Information Gathering

Description

Comprehensive guide for researching topics, gathering information from the web, local files, and knowledge base, synthesizing findings, and preserving learned knowledge for future use.

Triggers

• research
• find information
• look up
• summarize
• investigate
• analyze
• compare
• what is
• how does
• explain
• find out
• check
• verify
• learn about

Instructions

1. Research Strategy Selection

Choose your approach based on the information source:

Knowledge Base First (Always Start Here)

1. knowledge_search with the query — you may already know this from past tasks
2. If results are relevant and sufficient, use them — no need for external research
3. If partial results, use them as context and supplement with other sources

Local File Research

1. file_list to find relevant files (use pattern matching: *.py, *.md)
2. file_read for specific files — use line ranges for large files
3. For code analysis, self_read_source gives better context for agent source files
4. For project-wide searches, shell_execute with grep/find for complex patterns

Web Research

1. browser_navigate to authoritative sources first (official docs, primary sources)
2. For general queries, start with a search engine (Google, DuckDuckGo)
3. browser_extract for text content, browser_read_semantic for structured overview
4. Cross-reference multiple sources when accuracy matters
5. For APIs, browser_get_network can reveal data endpoints directly

2. Source Hierarchy

Prioritize sources in this order:

1. Primary sources — Official documentation, original papers, company blogs, government sites, specification documents
2. Authoritative aggregators — Wikipedia (for overview), MDN (for web tech), Python docs (for Python)
3. Community knowledge — Stack Overflow (verified answers), GitHub issues, technical blogs with code examples
4. General web — News articles, forum posts, social media (lowest reliability)

3. Research Workflows

Fact-Checking / Verification

1. knowledge_search → check if we already know this
2. browser_navigate to the most authoritative source
3. browser_extract → get the relevant passage
4. If conflicting claims: check 2-3 additional sources
5. Report findings with source attribution
6. knowledge_write to save verified facts for future use

Topic Deep-Dive

1. knowledge_search → existing knowledge
2. browser_navigate → overview article (Wikipedia, docs homepage)
3. browser_read_semantic → structured overview of the topic
4. Identify key subtopics from the overview
5. browser_navigate to each subtopic's authoritative source
6. browser_extract → detailed information per subtopic
7. Synthesize findings into a structured summary
8. knowledge_write to preserve the research

Competitive / Comparison Research

1. Identify the items to compare
2. For each item:
   a. browser_navigate to its official site
   b. browser_extract → features, pricing, specs
   c. browser_navigate to review/comparison sites
3. Build a comparison matrix from collected data
4. Present findings with clear differentiators

Current Events / Recent Information

1. browser_navigate to news sources
2. browser_extract for article content
3. Check publication dates — prioritize the most recent
4. Cross-reference across 2-3 sources for accuracy
5. Distinguish between confirmed facts and speculation

Technical Documentation Lookup

1. browser_navigate directly to the docs site (e.g., docs.python.org)
2. Use the site's search if available:
   a. browser_get_elements to find the search input
   b. browser_type the query
   c. browser_get_elements to find results
3. browser_extract the relevant documentation section
4. If the docs are paginated, follow links to subpages

4. Information Synthesis

Lead with the Answer

• State the conclusion or answer first
• Then provide supporting evidence and sources
• Don't narrate the research process ("First I searched..., then I found...")

Handle Uncertainty

• If confident: state directly
• If likely but not certain: "Based on [source], this appears to be..."
• If conflicting: "Sources disagree — [source A] says X while [source B] says Y"
• If unknown: "I couldn't find reliable information on this"

Cite Sources

• When reporting facts from the web, mention where they came from
• For critical decisions, provide the URL so the user can verify
• Don't over-cite obvious/common knowledge

Structured Output

• For comparisons: use a table or structured list
• For explanations: start simple, add detail as needed
• For summaries: lead with key points, details below
• For data: present the most relevant subset, offer the full set

5. Preserving Knowledge

After completing research, save valuable findings:

knowledge_write:
  path: "learned/research/{topic}.md"
  content: |
    # Topic Name
    
    ## Summary
    Key findings in 2-3 sentences.
    
    ## Details
    The full research findings.
    
    ## Sources
    - [Source 1](url)
    - [Source 2](url)
    
    ## Last Updated
    YYYY-MM-DD

When to save:

• Factual information the user might ask about again
• Technical documentation lookups for tools/services the user uses
• Comparison research that took significant effort
• Any finding that required multiple sources to verify

When NOT to save:

• One-off queries the user won't revisit
• Information that changes rapidly (stock prices, weather)
• Content that's trivially searchable

6. Common Pitfalls

• Don't guess — if you're not sure, research it rather than stating potentially wrong information
• Don't over-research — for simple factual questions, one authoritative source is enough
• Don't forget the knowledge base — always check knowledge_search first before going to the web
• Don't scrape paywalled content — if a page requires login/payment, tell the user
• Don't present search engine snippets as facts — navigate to the actual page and read the full content
• Don't ignore dates — information from 2020 may be outdated in 2026; check recency

Verify

• Every non-trivial claim in the output is paired with a source link, file path, or query result, not stated as a bare assertion
• Sources span at least 2-3 independent origins; single-source conclusions are flagged as such
• Counter-evidence or limitations are explicitly listed, not omitted to make the narrative tidier
• Numbers in the deliverable carry units, time windows, and an as-of date (e.g., '$1.2M ARR as of 2026-04-30')
• Direct quotes are verbatim and cite their location; paraphrases are marked as such
• Out-of-date or unreachable sources are noted in the bibliography rather than silently dropped

Notes

The knowledge base (knowledge_search / knowledge_write) is your institutional memory. Use it aggressively — every substantial research task should leave behind a knowledge artifact for future sessions. The agent across sessions only remembers what was explicitly saved to the knowledge base or task memory.