deep-research

What This Skill Does

The Deep Research skill empowers OpenClaw agents with advanced research capabilities using Google Gemini’s deep research agent architecture. It provides an asynchronous, reliable way to perform complex, multi-step inquiries that require synthesizing information from the web alongside your own local documentation. Unlike standard chatbot interfaces, this skill leverages RAG (Retrieval-Augmented Generation) grounding, allowing the model to look at specific local files—such as technical specifications, meeting notes, or research papers—to provide highly context-aware answers. It includes robust session management, cost transparency via a dry-run feature, and automated ephemeral file store handling to ensure privacy.

Installation

To integrate this skill into your environment, use the OpenClaw management CLI:

clawhub install openclaw/skills/skills/24601/agent-deep-research

Ensure that you have set your API credentials (GOOGLE_API_KEY, GEMINI_API_KEY, or GEMINI_DEEP_RESEARCH_API_KEY) in your environment variables. The skill does not require the Gemini CLI to be installed; it interacts directly with the Google Generative AI SDK.

Use Cases

• Technical Documentation Synthesis: Point the agent at a directory of SDK docs to generate an updated integration guide for a new API version.
• Market Analysis: Conduct deep web research on competitor offerings while grounding the output against internal strategy memos stored locally.
• Long-Form Reporting: Offload the heavy lifting of gathering, verifying, and formatting information from multiple sources into a structured, reliable document.
• Cost-Effective Prototyping: Use the dry-run feature to estimate token consumption before committing to expensive research tasks.

Example Prompts

1. "Perform a deep research session on the impact of quantum computing on modern encryption, grounding your answer in the provided ./docs/security-research folder."
2. "Research the latest best practices for React 19 performance optimization and compare these against my local project files in ./src/components, then summarize the findings in report.md."
3. "Estimate the cost of running a deep research session about the current regulatory status of AI in the EU using the files in ./legal-context/."

Tips & Limitations

• Security: The tool has built-in filters to prevent uploading sensitive files like .env, .pem, and node_modules. Always review the files you include in your --context flag.
• State Management: The skill creates a .gemini-research.json file to manage session state. If a research task is interrupted, you can use the state.py utility to clean up orphaned stores.
• Non-Interactive Environments: In non-TTY environments (like CI/CD pipelines), confirmation prompts are bypassed. Ensure you have defined your scope correctly to avoid unexpected costs.