This orchestrates multi-source research by spinning up subagents for different subtopics, having them save findings to files, then synthesizing everything into cited reports. You create a research plan, delegate 2-5 focused questions to parallel subagents (each doing 3-5 searches max), then read their findings files and combine them. It's structured around file-based coordination rather than direct returns, which keeps things organized in a dedicated folder. The workflow is methodical: plan, delegate, synthesize. Works well when you need to compare options or investigate something with multiple angles, though the overhead probably isn't worth it for simple lookups.
npx skills add https://github.com/langchain-ai/deepagents --skill web-research