prd — A PRD whose "Why" is fully cited

Generate a draft PRD from an Ideas/ file. The differentiator: the Why section contains only claims that exist in the graph, each cited — and the Open questions section shows engineering exactly what's still unvalidated, pulled from Research backlog.md. A PRD that admits what it doesn't know.

Trigger

/prd Ideas/Idea3 Context mode (Listening Modes).md

Structure of the draft

1. Problem statement — the When/I want/So that, expanded into prose from the linked Pain and its Signal (verbatim quote included — a real user's sentence beats ten bullets).
2. Why now / evidence — every linked Evidence and Signal, one line each: claim + source + Level (analyst doc — codes fine) + freshness note where the 3-month rule bites. Nothing uncited enters this section, period.
3. Who it's for — the persona(s) whose Pains link here (run the logic of /feature-panel briefly: fit, indifferent, conflict — including any persona this could hurt).
4. Scope sketch — reasonable functional bullets from the idea; clearly labeled as proposal, not research (this is the one section allowed to go beyond the data, and it says so).
5. Success metrics — from the idea's Metric to watch + how to verify via /analytics-sync; baseline value if an internal Evidence already holds one.
6. Open questions & risks — matching open rows from Research backlog.md + anything Low/Very-low Confidence in the scoring block, each with the cheapest next validation step (interview round via /interview-guide vs analytics query vs desk research).
7. Out of scope / won't do — anything the data actively argues against (e.g. a guardrail like "she won't pay a separate add-on charge [Price sensitivity]").

Workflow

1. Read the Idea + linked sources + persona sections + backlog rows + Product Context.md.
2. Draft; show it; iterate with the user.
3. On approval, save as PRDs/<Idea name> <YYYY-MM-DD>.md (create the folder if needed). Offer /export ticket for the implementation-sized version.

Rules

• The Why section is load-bearing: if an Idea's grounding is thin (L2 or less), the PRD's own header says so — "Draft based on desk research only; no interview validation yet" — instead of dressing up thin evidence as conviction.
• Stamp the draft with its data snapshot date; stale sources carry the refresh recommendation into the PRD.
• A PRD is a proposal document, not data — it never feeds back into Evidence/Signals.