Back to library

PRD Stress Tester

A PRD reviewed only by its author ships with the author's blind spots baked in. This skill reads your PRD as four different critics simultaneously — a sceptical senior engineer, a confused new engineer, a frustrated QA lead, and a product-savvy stakeholder — and surfaces every gap before the team sees it.

---

Context

This skill does not rewrite your PRD. It interrogates it. The output is a structured set of findings you can act on. Think of it as a red team for your product thinking.

---

Step 1 — Receive the PRD

Ask the user to paste the full PRD or feature spec. Accept any format.

Step 2 — Read as four critics

Critic 1: The Sceptical Senior Engineer

"What would I need to know to build this that isn't written here?"

Look for: missing behaviour specs, undefined data sources, missing performance requirements, ambiguous acceptance criteria, missing error states.

Critic 2: The Confused New Engineer

"What would I misunderstand and build wrong?"

Look for: undefined jargon, multiple valid interpretations, contradictions, implicit assumptions, missing context on why.

Critic 3: The Frustrated QA Lead

"What can I not test with what's written here?"

Look for: missing test cases, non-deterministic outputs without thresholds, unaddressed edge cases, no definition of failed output.

Critic 4: The Product-Savvy Stakeholder

"What business assumption is this built on that might be wrong?"

Look for: unvalidated user problems, scope larger than justified, missing success metrics, undefined launch criteria.

Step 3 — Classify every finding

Each finding gets: ID, Critic, Section, Issue, Impact, Severity (Blocker/High/Medium/Low), and Fix needed.

Step 4 — Generate the stress test report

Summary with counts by severity, readiness verdict (🔴 NOT READY / 🟠 NEARLY READY / 🟢 READY WITH NOTES), all findings grouped by severity, top 3 things to fix right now, and what the PRD does well.

Quality check before delivering

All four critic perspectives have been applied
Every finding has a severity, impact, and specific fix action
Blockers are clearly separated from lower-severity items
The "What this PRD does well" section is present and genuine
The readiness verdict is clearly stated
Top 3 fixes are actionable
Suggested next step: Fix all blockers and high findings, then run aipm-acceptance-criteria to generate testable criteria for the AI outputs this spec describes.