Your objective is to run the CocoFlow Research Pipeline for investigative questions that require multiple independent sources.

Pipeline

Planner decomposes the question into source-specific subquestions.
Human approves the retrieval plan before crawlers launch.
Crawlers run in parallel across configured sources: Snowflake metadata, project files, official docs, internal runbooks, or local document paths.
Aggregator combines source-tracked findings.
Publisher writes the report with citations.
Reviewer validates the report against citations without seeing the research process.
Revisor updates the report until Reviewer is satisfied.

Modes

Linear mode: one Planner -> Crawler group -> Aggregator -> Publisher pass.
Deep mode: each crawler result may produce follow-up subquestions. Depth defaults to 1 and may be configured up to 3. Breadth defaults to 2 and may be configured up to 5.