version: "1.0.0" name: auto-researcher description: > Proactively researches any topic by searching, synthesizing, and presenting well-sourced information — without waiting to be asked. Triggered when the user asks about current events, wants in-depth learning, says "what's the latest on...", "research X for me", "find out about...", "give me a briefing on...", or when the task would benefit from up-to-date external information that might be beyond Claude's training cutoff. Also triggers mid-task proactively when live data, prices, specs, or recent developments would meaningfully improve the answer. Output is structured: a key finding, 3-5 supporting evidence points with sources, any conflicting views, known limitations, and 2-3 further reading links. Source quality is tiered (peer-reviewed primary sources down to anonymous forums), and confidence levels are flagged inline for every major claim. Ideal for competitive research, technical deep-dives, market analysis, and any scenario where fresher information beats remembered knowledge. Also for: keeping a briefing current during a long session, fact-checking claims mid-conversation, and tracking how a topic evolves over time.

Auto Researcher Skill

Core Behavior

Don't wait to be asked to search. If current information would improve the answer, search automatically and synthesize before responding.

Research Process

1. Query Planning (do this before searching)

Break the topic into 3-5 targeted search queries:

• Start broad, then narrow
• Include date modifiers for recent info ("2025", "latest", "recent")
• Search for both facts AND expert opinions
• Search for counterarguments or limitations

2. Execute Searches

• Run multiple searches in parallel when possible
• Fetch full pages for key sources, not just snippets
• Prioritize: primary sources > reputable publications > secondary sources
• Note publication date of each source

3. Synthesis

Structure findings as:

• Key Finding: 1-sentence answer to the core question
• Supporting Evidence: 3-5 data points with sources
• Conflicting Views: Any disagreement among sources
• Limitations: What's unknown or uncertain
• Further Reading: 2-3 best sources for deep dive

4. Source Quality Tiers

• ✅ High: Peer-reviewed, official docs, primary data, major news outlets
• ⚠️ Medium: Industry blogs, secondary analysis, aggregators
• ❌ Low: Forums, anonymous sources, undated content

Research Templates

Competitive Research

1. Search "[Company] latest news 2025"
2. Search "[Company] product updates"
3. Search "[Company] vs competitors"

Technical Research

1. Official documentation first
2. Search "[technology] best practices 2025"
3. Search "[technology] limitations problems"

Market Research

1. Industry reports and statistics
2. Recent news and trends
3. Expert opinions and analyst views

Output Format

Always cite sources inline. Flag confidence level (High/Medium/Low) for key claims. Note knowledge cutoff vs. searched information distinction.

 
---
 
## Session Fetch Cache (ETag/304 Revalidation)
 
When re-checking the same URLs across research sessions, use HTTP conditional requests to avoid redundant fetches. This is particularly valuable in iterative research that revisits the same documentation, competitor pages, or reference sources.
 
**How it works:**
1. On first fetch, save the `ETag` or `Last-Modified` response header alongside the cached content
2. On subsequent fetches, pass `If-None-Match: <etag>` or `If-Modified-Since: <date>` in the request headers
3. If the server returns HTTP 304 Not Modified, the cached content is still current — use it without re-fetching
 
**Practical guidance for research workflows:**
- Save `{topic}-page.md` + `{topic}-page.md.etag` together when caching a source
- Before re-fetching a known URL, check for a `.etag` file and use conditional request if found
- A 304 response = content unchanged; use the cached version, skip the new fetch
- Not all servers send ETags. If absent, use `Last-Modified`, or fall back to full fetch
- Credit-aware workflows: check ETag first, only consume credits when content actually changed
 
**When to use:** Long-running research projects that monitor the same sources over time; competitive tracking workflows; documentation research that needs daily freshness without redundant full fetches.
 
See also: `firecrawl-scrape` Session Fetch Cache section for the CLI-level implementation.
 
---
 
## Output Format