AI agents for UCD teams. Built to a standard.

User research, service design, interaction design, content, business analysis, accessibility, and performance.

Each agent drops into your own AI tool, works one stage at a time, and stops for your yes.

Every one is built with the same five-question method — published in full.

See the nine agentsHow they're designed

Or get all nine in the UCD Team bundle

Works in Claude Code and Cursor · Your own AI keys · Nothing to host · WCAG 2.2 AA built in

The problem

Most agent projects skip the questions that would have caught failure early.

  • Teams build agents before they know what they are for.

  • Nobody says when a person must step in.

  • Nobody can prove the agent works.

How every agent is designed

Every agent answers five questions before it is built. You can read every answer.

Read the full methodDownload the template (free)

Seven components

What goes in the zip once the five questions are answered.

FileAnswersContents
instructions.md1, 2Role, voice, tone, refusals, two worked examples
skills/1One file per skill: inputs, steps, outputs, evidence rule, gate
tools.md3Every connection with read/write scope and fallback
manifest.md3Which skill uses which tool. No gaps, no guesses
guardrails.md2, 4Each risk, its control, its owner, sign-off
memory.md2What is kept, where, how long, who agreed it
evals/5Test inputs, expected outputs, pass marks, last score

Nine agents

Each package is built to a named standard. All nine are available on Gumroad.

How the package works

No hosting. Your keys. Your tool.

Download

Get it from Gumroad

Buy once and download the zip straight away.

Install

Drop it in your project

Unzip into a repo and open your AI tool.

Run

Approve each stage

Say the trigger phrase. The agent stops for your yes.