Autopilot Research

Track any topic across social platforms. Raw data is captured and stored. KB documents are compiled and kept current via periodic crawl jobs.

This skill automates the full research lifecycle: initial deep dive, evidence storage, and ongoing monitoring with automatic KB updates.

---

When to Use This Skill

Use autopilot-research when:

• The user wants ongoing monitoring of a topic, brand, or competitor
• Research needs to stay current over weeks/months (not a one-shot question)
• Raw evidence matters — the user wants to see source data, not just summaries
• The topic spans multiple platforms and needs cross-platform synthesis

Do NOT use when:

• A quick one-time analysis suffices (use run_analysis instead)
• The user only needs to search for specific posts (use search_content)
• The knowledge base already has fresh compiled truth (use recall first)

---

Workflow

Step 1: Check Existing Knowledge

Always check first:

recall({ query: "[topic keywords]" })

If recent compiled truth exists (< 7 days), present it and ask if refresh is needed.

Step 2: Run Initial Research

Use the decomposition pipeline for comprehensive multi-platform research:

run_analysis({
  question: "[research question]",
  platforms: ["reddit", "twitter", "hackernews", "news", "youtube"]
})

Available platforms (12 total):

Platform	API	Auth Required	Cost
reddit	Exa + Reddit OAuth	—	Existing
twitter	Exa	—	Existing
hackernews	Algolia HN Search	None	Free
bluesky	AT Protocol	None	Free
youtube	YouTube Data API v3	YOUTUBE_API_KEY	Free (100/day)
tiktok	ScrapeCreators	SCRAPECREATORS_API_KEY	$47/mo shared
instagram	ScrapeCreators	Same key	Same
threads	ScrapeCreators	Same key	Same
producthunt	GraphQL API v2	PRODUCTHUNT_ACCESS_TOKEN	Free
github	GitHub API	GITHUB_TOKEN	Existing
news	Exa	—	Existing
xiaohongshu	Visual extraction	—	Existing

Platforms without required API keys are automatically skipped. Exa fallbacks are used for YouTube and TikTok when their API keys are missing.

The system will:

1. Break the goal into platform-specific subtasks
2. Run independent platform searches in parallel
3. Store all raw search results in the evidence database
4. Compile findings into a KB document with source-attributed citations

Step 3: Set Up Periodic Tracking

After initial research, create a monitored entity for automatic updates:

manage_entities({
  action: "create",
  name: "Topic Name",
  type: "topic",
  keywords: ["search terms"],
  config: {
    research_enabled: true,
    research_interval_hours: 168,
    search_query: "the research question",
    kb_path: "research/topic-slug",
    max_sources_per_refresh: 30,
    lookback_days: 7
  }
})

The scheduler will automatically:

• Crawl configured platforms at the specified interval (default: weekly)
• Store new raw results with deduplication
• Route findings to the best KB folder/document using LLM classification
• Merge new findings into existing documents (or create new ones)
• Preserve source citations with platform, author, date, and engagement
• Update the document's Data Quality section

Step 4: Review Updates

Check the KB document for updates:

recall({ query: "topic keywords", path: "research/topic-slug" })

Check raw evidence:

search_content({ query: "topic keywords", search_mode: "knowledge" })

---

Tracking Configuration

Interval Options

Interval	Hours	Use Case
Daily	24	Fast-moving topics, crisis monitoring
Every 3 days	72	Active product launches, trending topics
Weekly	168	Standard monitoring (default)
Biweekly	336	Stable topics, competitor tracking

Adjusting Tracking

Update an entity's config to change crawl behavior:

manage_entities({
  action: "update",
  entity_id: "[id]",
  config: {
    research_interval_hours: 72,
    max_sources_per_refresh: 50
  }
})

Pause tracking:

manage_entities({
  action: "update",
  entity_id: "[id]",
  config: { research_enabled: false }
})

---

KB Folder Routing

New findings are automatically routed to the best location in your knowledge base:

1. Merge — If an existing document already covers the topic, new evidence is merged in
2. Create in folder — If a matching folder exists (e.g., pain-points/), a new doc is created there
3. Default — Falls back to research/{slug} if nothing matches

This means research compounds across crawl cycles instead of creating duplicate documents.

---

KB Document Structure

The compiled KB document follows this structure with inline source citations:

# Research: [Topic]
*Last updated: [timestamp]*

## [Theme 1]
> "Key finding or quote from the source"
> — [Platform] · @author · 2026-04-15 · 142 upvotes
> [Source](url)

Additional analysis and context...
*8 sources · Last updated: 2026-04-15*

## [Theme 2]
...

## Data Quality
- Total sources: N
- Platforms covered: [list]
- Date range: [oldest] to [newest]
- Last crawl: [timestamp]

Findings are grouped by theme (not by platform). Each update merges new evidence, preserves existing citations, removes outdated info, and maintains document quality.

---

Anti-Patterns

1. Setting up tracking without initial research. Always do a deep dive first.
2. Tracking too many topics. Each tracked entity runs periodic searches — be selective.
3. Ignoring the Data Quality section. If source counts are low or dates are stale, the topic may need manual attention.
4. Using daily intervals for stable topics. Weekly is sufficient for most use cases.

---

Example: Complete Autopilot Setup

User: "Monitor what people think about Cursor AI"

1. Initial Research

run_analysis({
  question: "What do developers think about Cursor AI? Focus on productivity, pricing, and comparison with alternatives.",
  platforms: ["reddit", "hackernews", "twitter", "github", "youtube", "news"]
})

2. Review Results

The system creates a KB document at research/cursor-ai-developer-sentiment with platform-annotated sections.

3. Auto-Tracking Enabled

The system automatically creates a tracked entity. Weekly crawls will update the KB document with fresh evidence.

4. Check Updates (next week)

recall({ query: "cursor ai developer sentiment" })

The document now includes both the original research and new findings from the weekly crawl, with updated source counts and timestamps.