to-prd
To PRD
Turn settled choices into one reviewable PRD. The result is an approved specification that to-tasks can translate into executable slices.
Boundary
Receive .ai-workflow/work/<feature-slug>/decision-record.md with status ready-for-prd. Produce .ai-workflow/work/<feature-slug>/prd.md first with status draft, then with status approved after the user accepts it.
Read shared/references/workflow-stage-routing.md before writing. This step preserves settled choices. It does not open fresh architecture choices, prescribe test mechanics, select files, or decide an implementation model or effort level.
Do not call a panel, a runner, models-consensus, design-gate, security-gate, or test-lens. If the user explicitly wants additional opinions, direct them to invoke models-consensus. When the input has a material missing decision or a conflict with repository facts, return the exact question to interview-me. Do not guess or conduct a second interview here.
Work
-
Confirm that the decision record status is
ready-for-prd, then read it with the relevant code, domain glossary, and prior architecture decisions. A draft record returns tointerview-me. Respect the already decided choices. If repository facts contradict a settled choice, describe the conflict and return it tointerview-mebefore producing an approvable PRD. -
Write the PRD draft at the stated path. Start it with
# PRD: <feature name>and**Status:** draft. -
Include these sections: