research

The research skill is a specialized agent capability designed to ensure high-accuracy information retrieval and fact-checking. It is intended for developers, researchers, and technical writers who require verifiable data rather than generic summaries. By implementing a strict verification protocol, it prevents the common pitfall of presenting LLM-generated snippets or search engine summaries as absolute facts. The skill forces the agent to move beyond superficial search results by requiring a 'fetch-read-verify' loop, which acts as a safeguard against misinformation in technical tasks.

• Performs deterministic WebFetch operations to retrieve raw content from provided URLs.

• Enforces strict verification of claims against primary source documentation to ensure accuracy.

• Supports multi-tool integration, including WebSearch, Kagi summarizer, and Bash-based inspection tools for deep-dive investigations.

• Facilitates repository-level research by allowing subagents to clone and inspect documentation or source code repositories directly.

• Encourages explicit error reporting when information is behind paywalls, returns 404 errors, or exhibits internal contradictions.

• Always utilize WebFetch to inspect full pages before finalizing any findings or summaries.

• When in doubt, explicitly state that a claim cannot be verified rather than guessing or inferring from incomplete data.

• Prioritize direct quotations from official documentation, RFCs, or project-specific references over third-party blog posts.

• Use this skill during complex technical research where citation accuracy is critical for system design or debugging.

• Avoid presenting snippets as facts without verifying the original context; use the tool as a verification agent, not just a search proxy.

• Integrate with repository tools and grep to perform cross-reference checks between online documentation and actual implementation patterns.