PRD Focus Group
Pressure-test a Product Requirements Document before you write a line of code. Give it a PRD; it imagines the shipped product, runs 20 fixed user personas through it, convenes a 5-person PM panel to critique and converge, and hands back a prioritised set of fixes.
A Claude Skill · runs on Claude.ai, Cowork, or Claude CodeWhat it does
4-STAGE PIPELINEImagine the product
Reconstructs the product as if it had already shipped exactly as your PRD specifies — the surface real users would actually meet.
Run 20 user personas
Varied backgrounds, economic status, tech-literacy, and accessibility needs move through a configurable time arc, each producing an honest friction / delight log.
Convene a 5-person PM panel
A traditionalist, a disruptor, a design-led PM, a growth/metrics PM, and a platform/technical PM read the findings, critique each other, and converge on prioritised fixes.
Emit three artifacts
A findings report, a redlined PRD, and the raw persona logs — decisions handed back to you, not invented for you.
The cast is fixed and inspectablereproducible
Every persona and PM lives in References/, so runs are repeatable and you always know exactly who is in the room:
- Traditionalist PM
- Disruptor PM
- Design-led PM
- Growth / metrics PM
- Platform / technical PM
- + 20 user archetypes
Get it
2 WAYSClaude.ai / Cowork
No clone needed — grab the packaged skill and upload it.
- Open the latest release.
- Download
prd-focus-group.skillfrom the assets. - Upload it in the Skills panel.
Claude Code
Clone the repo and drop it into your skills directory:
git clone https://github.com/jamesh-Pivotal/FocusGroup_PRD.git
cp -r FocusGroup_PRD ~/.claude/skills/prd-focus-group
Claude Code discovers it via SKILL.md at the folder root.
Run it
JUST ASKOnce installed, hand Claude your spec and name an arc:
Run the PRD focus group on this spec. Use the retention arc over 30 days.
[attach or paste your PRD]
Arcs: retention, first-impression, feature-walkthrough, or custom. On Cowork / Claude Code the personas run as real parallel sub-agents (more rigorous); on plain Claude.ai they run sequentially in-context. It always degrades gracefully, so it never hard-fails.