research-tool

What This Skill Does

The research-tool is a powerful OpenClaw skill designed to bridge the gap between your local agent and the live internet. By leveraging OpenRouter's massive model library, this tool allows your agent to perform deep web searches, synthesize information from multiple sources, and provide cited answers to complex queries. Whether you are performing market research, debugging obscure API documentation, or fact-checking live news, this skill acts as your agent's window to the real world.

Installation

To begin using the research-tool, you must first have the OpenClaw environment set up. Install the tool directly via Cargo:

cargo install openclaw-search-tool

Alternatively, if you are utilizing the ClawHub ecosystem, you can install it via: clawhub install openclaw/skills/skills/aaronn/openclaw-search-tool. Ensure your OPENROUTER_API_KEY environment variable is exported in your shell configuration; this is required for the tool to authenticate and access the various LLMs supported by OpenRouter.

Use Cases

• Technical Documentation Lookups: Retrieve the latest API limits, syntax, or breaking changes for modern frameworks like React, Rust, or cloud services.
• Market Research: Gather and summarize competitive data, recent news articles, and industry reports.
• Fact-Checking: Verify information by instructing the agent to cross-reference multiple reputable URLs.
• Complex Reasoning: Use the --effort flag to have the model break down multi-faceted problems that require synthesizing data from diverse sources.

Example Prompts

1. "research-tool 'What are the current limitations of the Anthropic API compared to the latest GPT-5 release?'"
2. "research-tool --effort high 'Summarize the recent security patches for the Linux kernel and explain how they impact containerized environments.'"
3. "research-tool 'Find the latest documentation on integrating OpenClaw sub-agents and provide a code snippet example.'"

Tips & Limitations

• Sub-Agent Execution: Because the research process involves browsing and multi-step reasoning, it can be time-consuming. Always run this skill within a sessions_spawn command. This ensures your main terminal session remains responsive while the agent processes the query in the background.
• Do Not Set Timeouts: The most common user error is applying a rigid timeout flag to the exec command. Because the tool performs live web interactions, it requires significant time. Use yieldMs to background the task rather than forcing a premature termination.
• Model Versatility: While it defaults to openai/gpt-5.2:online, you can swap the --model flag to use any model on OpenRouter that supports the :online suffix, allowing you to choose between reasoning-heavy models or faster, more efficient ones depending on your task priority.