blindspots

Blindspots

A Claude Code plugin that gives you synthetic users for your product. They use it like real people would — and find things you might not otherwise notice.

How it works

You install the plugin and type /dogfood. First run walks you through setup — asks about your product, generates config, creates personas. Then it picks a persona and starts using your product in character.

Blindspots works with webapps, CLI tools, plugins — anything you can run locally.

From there, four modes let you ask different questions about your product:

Does it work? /dogfood

Your persona tests the product against your specs. When it finds something that doesn't match — a broken flow, a missing feature, a wrong behavior — it writes a failing test, fixes the code, and opens a PR. Then it keeps going.

Can someone actually use this? /user-trial

The persona gets no specs, no docs, no source code — just a browser and a URL. It explores on its own and writes down what it thinks your product does. Then Blindspots compares those observations against your actual specs. The gaps between what you built and what they perceived — those are your blind spots.

Which version is better? /experiment

Run trials against two variants of your product in parallel — your main branch vs. a redesign, v1 vs. v2, localhost:3000 vs. localhost:3001. Same personas, both versions. Produces a scorecard and a ship/no-ship verdict. Test ideas without releasing them to real users.

Why? /interview

Talk to a persona directly. Ask what they expected before a user trial, how they felt after one, or what they think of your onboarding. Synthetic user research, on demand.

Install

# Add the marketplace (one-time)
claude plugin marketplace add btucker/blindspots

# Install the plugin
claude plugin install blindspots@blindspots

Setup

1. Run /dogfood from within Claude Code.
2. Follow the prompts

Config

Running any command for the first time triggers interactive setup. It reads your project, asks a few questions, and generates .blindspots/config.md and .blindspots/personas.md. You can also run /setup directly to regenerate.

A generated config looks something like:

## Setup
npm run dev &

## Start
http://localhost:3000

## Specs
- SPECS.md

## Start tells the persona how to begin — a URL for webapps, a command for CLI tools, whatever makes sense for your product.

Optional config sections

Section	Purpose
## Explore	Features and flows to test
## Diagnose	Where to look when things break
## Test	Test framework and conventions

Personas

Generated automatically on first run from your project context and web research about your target audience. Saved to .blindspots/personas.md — check it in, edit it, make them yours.

/dogfood, /user-trial, and /experiment pick a random persona unless you pass --persona "name". /interview shows you the list and lets you choose.