prd-studio

Automating PRD writing and reviewing is achieved by orchestrating isolated subagents that cycle through write, review, fix, and review phases until approval or a maximum of three iterations, tracking progress via file-based status channels.

The maximum number of review iterations for an automated PRD workflow is three cycles. The orchestration loops through write, review, and fix phases until the reviewer subagent approves the PRD or reaches this iteration limit.

Subagent isolation prevents context leakage during PRD generation by using Task-driven subagents for each phase. This ensures the writer and reviewer operate independently without bleeding context, producing clean artefacts like status and review reports.

Generating multiple PRDs simultaneously is supported through multi-PRD scenario orchestration. The workflow manages both single and multiple PRD generation by cycling isolated writer and reviewer subagents to produce and approve documents independently.

Specific file structures are required to track PRD workflow status and reports. The orchestration relies on a file-based state channel, specifically maintaining workflow/status.md, workflow/prd.md, and workflow/review-report.md to manage phases and artefacts.