Research Deep

This automates deep research by spinning up parallel agents for each item in your outline. Point it at an outline.yaml file and it launches independent web search agents in the background, each writing structured JSON output to disk. The setup is clever: it resumes from where you left off, batches work for your approval, and validates output against field definitions. You'd use this when you need to research dozens of similar items (like competitive products or market segments) and want consistent structured data without babysitting each query. The task output is disabled since agents write directly to files, which keeps your terminal clean during long runs.