Draft & critique a PRD
Turn a rough PRD into something that survives exec scrutiny — and find the weak assumptions before someone else does.
A PRD is only as strong as its weakest assumption. Use a model to play the skeptic before your stakeholders do. The flagship models reason hardest about hidden risk; a balanced model is plenty for a first tidy-up.
Claude Fable 5
Why this variant: The final pass for the PRD that actually goes to the board. It reads the whole document like a detail obsessive — reconciling numbers, owners, dates across sections — instead of sampling the highlights. Hand it the complete doc in one brief and skip the scaffolding; it burns tokens, so this is the last pass, not the first.
<role>You are the most detail-obsessed reviewer in the company. This PRD goes to the board tomorrow.</role>
<prd>
{paste the full PRD — don't trim it}
</prd>
<task>
Full detail audit. Find everything the room will catch: numbers that
don't reconcile, claims with no evidence, success metrics nobody owns,
sections that contradict each other, and the assumptions hiding inside
confident sentences. For each finding: where it is, why it matters, and
the exact fix. Be exhaustive — I'd rather filter your list than miss one.
</task>Claude Opus 5
Why this variant: Best when the PRD is messy and the stakes are high. The XML structure keeps the document and the task cleanly separated, so it critiques rather than rewrites — and you avoid re-paste loops that burn tokens.
<role>You are a seasoned CPO reviewing a PRD before it goes to the board.</role>
<prd>
{paste your PRD}
</prd>
<task>
Identify the three weakest assumptions. For each: why it's risky, the one
question an investor would ask, and how I'd de-risk it in a week. Be
blunt. Bullet points only — no summary of the PRD back to me.
</task>Claude Sonnet 5
Why this variant: The right default for a first clean-up pass — fix structure, tighten the problem statement, flag gaps. Cheaper and faster; escalate to Opus only if the critique feels shallow.
<prd>
{paste your PRD}
</prd>
Tighten this PRD. Return: (1) a sharper one-sentence problem statement,
(2) any section that's missing, (3) three sentences that are vague and
how to make them specific. Keep your answer under 200 words.GPT-5.6 Sol
Why this variant: Strong when the hard part is the logic of the bet, not the prose. Don't add "think step by step" — it reasons internally; just give it the goal.
Here is a PRD and the rationale for building it:
{paste PRD + rationale}
What single assumption, if false, would make this the wrong thing to
build? Give me that assumption, the cheapest way to test it this week, and
what result would tell us to stop.