research-deep

To automate deep research for outline items, you need a tool that reads outline.yaml, launches parallel web-search agents per item, and writes structured JSON outputs to a designated directory while validating field completeness.

Yes, resumable batch execution detects existing JSON files in the output directory and automatically skips completed items, allowing the research workflow to continue from the last unfinished outline item without duplicating effort.

You validate structured JSON outputs by defining field requirements in a fields.yaml file and running a validation script like validate_json.py to ensure each research item passes coverage checks before being marked complete.

The best way to run parallel web-search agents is processing outline items in batches with background execution, applying guardrails that request user approval before moving to the next batch of research tasks.

Yes, you need an outline.yaml file containing research topic metadata and the list of items to investigate, plus a fields.yaml file defining output structure, to resolve item metadata and generate structured JSON properly.