| name | prd-writing |
| description | Use when running /business:prd, after a Vision.md exists — derives a product-level epic list from the vision's goals. Philosophy adapted from BMAD-METHOD's PM-stage discovery. |
PRD Writing
A PRD sits between the vision (why/for whom) and feature specs (what a single feature does).
Its job is to decompose the vision's goals into epics — coherent clusters of capability —
at a level still too coarse to implement directly. Each epic becomes a candidate slug for a
future /spec:spec run.
Process
- Read the vision's goals first. Every epic should trace to at least one vision goal; an
epic with no goal behind it is scope creep at the product level.
- Decompose into epics, not features or tasks. An epic is "capability a user gains," not
"screen we'll build" or "function we'll write" — those are UX phase and engineering phase
concerns respectively.
- Prioritize honestly. Must/Should/Could, and be willing to put things in Could — a PRD
where everything is Must isn't prioritized at all.
- Note constraints, not solutions. Business, legal, or timeline constraints belong here;
how to satisfy them technically belongs to
/business:architecture or /spec:plan.
Boundary with feature specs
Do not write functional requirements here — that's /spec:spec's job, one feature at a time,
once an epic is picked up for implementation. If you catch yourself writing "the system MUST
...", that sentence belongs in a Specification.md, not the PRD.
Self-review checklist
Red flags
| Red flag | Why it matters |
|---|
| An epic reads like a functional requirement ("MUST validate email format") | Too granular — that belongs in a feature spec |
| An epic has no traceable vision goal | Likely scope creep slipping in without justification |
| Every epic is priority "Must" | Prioritization didn't actually happen |