Deep Research
Comprehensive research using multi-source evidence, synthesis, and citation structure.
Deep Research
Conduct comprehensive deep research on complex topics using Google's Deep Research agent with streaming progress updates.
When to Use
Use this skill when you need to:
- Research complex topics requiring multiple sources
- Generate comprehensive reports with citations
- Analyze and synthesize information from various angles
- Investigate technical concepts, industry trends, or detailed questions
Usage
Run the deep research script with your topic:
Optional: Save to a specific output file:
IMPORTANT: Wait for Completion
Deep research takes 3-10 minutes to complete. The script will:
- Stream thinking/progress updates in real-time
- Automatically retry and poll for results if needed
- Save the final report to a file when complete
DO NOT interrupt or check on the research while it's running. The script handles all retries and status checking internally.
Examples
Options
| Option | Short | Default | Description |
|---|---|---|---|
--output | -o | auto-generated | Output file path |
--timeout | -t | 600 | Maximum time in seconds (default: 10 min) |
--verbose | -v | false | Enable verbose debug logging |
--system | none | Custom system instruction |
Research Process
The script uses Google's Deep Research agent which:
- Starts an interaction: Creates a research session with the topic
- Streams thinking summaries: Real-time progress as the agent researches
- Delivers results: Final report with sources and citations
- Auto-recovers: Handles network issues and retries automatically
Output
The script generates a markdown report containing:
- Comprehensive research findings
- Source URLs for verification (in Sources section)
By default, reports are saved to /workspace/deep_research/ with auto-generated filenames.
IMPORTANT: Enhance Reports with Images
After the research completes, you MUST enhance the report with generated images:
- Generate a title image for the report header
- Generate 1-2 concept visualizations for key findings
- Insert the images into the report using sed
Example workflow after research completes:
Images should be saved to /workspace/generated_images/ and referenced with relative paths from the report.
Streaming Events
The script outputs JSON events for real-time UI updates:
{"event": "start", "topic": "..."}- Research started{"event": "thinking", "summary": "..."}- Progress update{"event": "content", "text": "..."}- Report content{"event": "complete", "report_file": "...", "sources_count": N}- Done
Requirements
GEMINI_API_KEYenvironment variable must be set- Python 3.10+ with
requestspackage installed
Notes
- Research typically takes 3-10 minutes depending on topic complexity
- The script blocks until completion - do not interrupt
- Results include sources for verification
- If research fails, the script will clearly indicate this