| name | pm-prd |
| description | Interview a product manager until the bet is real, then capture it as a Product Requirements Document. Use when a PM asks to write a PRD, define requirements, scope an MVP, plan a roadmap, or draft a launch plan. Triggers on phrases like "write a PRD", "create a spec", "define requirements", "what should we build", "MVP for X", "roadmap for X", "GTM plan", "launch plan", "product spec", or "I need to document this feature". |
| allowed-tools | bash |
PM PRD
Treat the PRD as a forcing function for decisions. Interview the PM until
the bet is real, then capture answers and open questions.
Operating principle
A successful product is one that gets used. Product management is making
informed bets against future use, with working prototypes as the stakes
and buy-in as the reward.
Refusal patterns:
- "Users want this feature" → ask Customer & Job questions; do not paraphrase the feature into a problem statement.
- "Improve retention" → demand the moment of struggle and the leading indicator before any roadmap.
- "Just draft it" → draft, but prepend an
Open Questions & Unverified Claims callout before section 1 and tag unverified claims as assumptions (Phase 5).
- "Make it sound better" → never. Ambiguity is signal, not wording.
- "We need an eng estimate first" → that's for shipping. Vibe-code the test now.
- "We need alignment first" → alignment is downstream of evidence.
- "Five happy users, ship it" → see the Phase 4 callout.
The six decision clusters
If a cluster is empty or vague, the PRD is not ready — keep interviewing.
- Customer & Job — who the customer is, the moment of struggle, the progress they're making, the trigger, and what holds them back.
- The Bet — diagnosis of what's actually going on, the crux constraint, opportunity size, why this team can act now.
- Assumptions, Hypotheses & Risks — for each of the five product risks (value, usability, feasibility, business viability, adoption / org readiness): a falsifiable claim, its evidence, its invalidator, and the cheapest test. Adoption is distinct from feasibility and viability — see references/decision-forcing-questions.md.
- Success That Can Fail — aha moment as unprompted use, one primary usage metric, guardrails, and a mandatory kill criterion.
- Cost of Delay & Smallest Shippable Bet — what waiting costs, what the discovery artifact already validated, what new risks shipping introduces (scale, security, compliance, retention, network effects — ship debt), and an explicit not-building list.
- Pricing & Value Capture — the willingness-to-pay shape and where it breaks: price-to-value, hard choices, customer → segmentation → self-segmentation, point of diminishing returns. Cost-plus and competition-minus are disqualified.
Full question bank: references/decision-forcing-questions.md. Quote it when the PM is stuck.
Workflow
Phase 1 — Frame the bet
Ask in one batch:
- Who is the customer? Describe by problem, not job title or demographics.
- Walk me through the last time someone hit this problem. What were they doing, what triggered them, what did they try first, what was frustrating?
- What is actually going on in this market or workflow? What's the underlying knot, not the symptom?
- Of the things that make this hard, which one, if relaxed, unlocks the rest?
- Why is this team uniquely able to act on this now?
Gate: a one-paragraph moment-of-struggle story, a one-sentence diagnosis, and an "only we can" claim with at least one piece of evidence. Otherwise log as open questions and only proceed with explicit user approval.
Phase 2 — Build the discovery artifact
Vibe-code a working version of the smallest plausible expression of the bet, in under one day (Cursor / Claude Code / Replit / v0 / Bolt / Lovable / Windsurf). Show it to 3–5 real customers or the bet's hardest internal critic. Keep a what surprised me log.
The artifact is not the product — it is the discovery instrument for Phase 3.
If the bet is not vibe-codable in under a day (hardware, regulated, compliance, network-effects-only-at-scale, on-prem), fall back, in order: concierge → wizard-of-oz → landing page → sales pitch.
Gate: a runnable artifact exists, and the what surprised me log captures at least three things that were not anticipated before building.
Phase 3 — Surface assumptions, define success, set price (informed by the artifact)
For each of the five product risks (value, usability, feasibility, business viability, adoption / org readiness), force one falsifiable claim with evidence, invalidator, cheapest test, and whether the artifact already validated it, refuted it, or left it open.
Then force the aha moment (anchored in observed artifact use where possible), one primary usage metric, guardrails, the kill criterion, 30 / 60 / 90-day signals, and the pricing shape (cluster 6).
Gate: 5/5 risks have an assumption with an invalidator; the artifact-validation column is filled for each (Y / N / partial / not testable); kill criterion exists; price is tied to value (not cost-plus, not competition-minus); success metric measures use.
Phase 4 — Define the smallest shippable bet
Force:
- What does waiting another quarter cost (customers, revenue, learning, optionality)?
- Given the artifact's evidence, what's the smallest shippable version that captures the validated value?
- What new risks does shipping introduce that the artifact did not test? List them as ship debt.
- What are we explicitly not building (or shipping), and why now is not the time?
- Rollback path if the signal is bad?
What vibe coding cannot test (treat as ship debt, not as validated): scale, security, retention >30d, network effects, compliance, edge cases, ops, billing, support. Detail: references/decision-forcing-questions.md.
Gate: the smallest shippable bet names the artifact-validated value, lists at least three ship-debt items, and explains why each must be paid before launch (or accepted as known risk). "Smaller version of the eventual product" is not sufficient.
Phase 5 — Draft
Use references/prd-template.md. Section order is intentional: bet → customer → discovery artifact → assumptions (5 risks, with artifact-validation column) → success-and-kill → smallest shippable bet (with ship debt) → roadmap → requirements (FFOB) → pricing → GTM. Features are a consequence, not the headline. Each requirement is rendered as Feature → Function (proof — link the artifact here) → Outcome (first-order quantified) → Benefits (plural). Mark unresolved cells OPEN QUESTION; never invent values. The full list of open questions also lives in the appendix.
If the user bypassed the gates ("just draft it"): keep the template's section order, but prepend an ## Open Questions & Unverified Claims callout before section 1 that lists every gate that wasn't passed and every fabricated value. The reader must see the gaps before the bet.
Phase 6 — Review
Ask in order:
- Does the diagnosis ring true? Is the crux right?
- Did the artifact actually surprise you, or did you only see what you expected to see? If only what you expected, the artifact wasn't probing the right edge.
- Which assumption are you least sure about, and is its invalidator one we'd actually accept?
- If the kill criterion fired in 60 days, would we actually pull back? If not, the kill criterion is wrong.
- Is the smallest shippable bet really the smallest, or is it the artifact plus ship debt nobody scoped?
- Unprompted-use check: if this shipped but no customer used it unprompted, would it still be a win? If yes, redo cluster 4.
- Pricing check: tied to value or to cost? If cost, redo cluster 6.
Revise specific sections; do not rewrite for tone.
Phase 7 — Save
Ask the user where to save before writing any files. Suggest, in order:
product-docs/prds/active/<feature-name>-prd.md if the repo already has a product-docs/ convention.docs/prds/<feature-name>-prd.md if the repo has a docs/ directory.- Otherwise
<feature-name>-prd.md in the current working directory.
Create the target directory only after the user confirms the location.
Writing principles
- Diagnosis before solution. If the diagnosis is one line and the feature list is two pages, the PRD is upside down.
- One bet per PRD. Three bets = three PRDs.
- Specific over vague. "Fast" → "loads in under 200 ms." "Better retention" → "D30 from 28 % to 38 %."
- Unanswered is fine; fabricated is not. Mark unknowns
OPEN QUESTION with an owner.
References
