| name | prd-generator |
| description | Use when an approved BRD must be translated into observable product requirements before FSD planning. |
PRD Generator
Define users, observable product behavior, rules, acceptance criteria, and outcomes between BRD exploration and technical FSD planning. Announce use before authoring.
When to Use
Use after an approved BRD exists and before /sc-plan. If no BRD is approved, return upstream. If a PRD already covers the topic, ask whether to revise it or create a separately scoped document.
Route
- For evidence collection, clarifying questions, required sections, and the compact shape, load PRD authoring.
- For story sizing, validation, storage, and downstream routing, load validation and handoff.
- Load the validation branch only after a draft exists.
Invariants
- Save durable output to
docs/prd/prd-<feature-name>.md; author from
.agent/templates/agentic-delivery/skeletons/PRD-Skeleton.md, then load only
a specific section on demand from the full PRD reference library.
- Trace requirements with qualified references such as
BRD-CCC#BREQ-001; do not duplicate source prose when an ID is sufficient.
- Preserve BRD business policy, scope, acceptance decisions, and domain language. The PRD must not invent technical implementation, schemas, databases, or internal architecture beyond known repository constraints.
- Every goal maps to at least one story or requirement; every story has specific, verifiable acceptance criteria and negative cases where risk warrants them.
- UI-bearing PRDs set
ui_delivery_profile, cover every named state or approved
N/A, record responsive/accessibility intent, and route the draft through
read-only /sc-ui before approval. HIGH_INTERACTION needs evidence matching
its runtime risk.
- Prefer vertical user-value stories that fit one focused implementation session.
- Record unresolved decisions as concrete
OPEN-* blockers. Do not disguise missing product decisions as implementation freedom.
Red Flags
- Writing from assumptions while approved source material or repository evidence is available.
- Horizontal layer stories, vague “works correctly” criteria, missing non-goals, or untraceable requirements.
- Technical design masquerading as product requirements.
- UI requirements without accessibility/responsiveness, or data/security/privacy requirements without failure and abuse cases.
- Proceeding to FSD while a scope- or policy-changing open item remains.
Integration
Upstream: /sc-explore, brainstorming, agentic-delivery, domain-modeling, codebase-design, and interface-design. Downstream: writing-plans, issue-workflow, and plan-verification. The pipeline is BRD → PRD → FSD/GOAL → execution; stop after saving and reporting the PRD unless further work is explicitly requested.
