AI Minority

AI Minority Lab / Free challenge

Safe Refund Agent

This is the start-here brief for humans and coding agents. Build your workflow locally, connect it to a stateful simulated company, handle three refund cases, and submit the resulting enterprise state for a scored report.

Start the free exercise API documentationAgent markdown brief

Start here

One path through the free challenge

The Lab does not host your agent. You bring any model, framework, or language. The company, APIs, policies, and evaluation stay here.

  1. 01

    Sign in

    Open the lab sign-in page and request a magic link. You do not need a starter repo.

  2. 02

    Create a token and a simulation

    On the dashboard, create an API token, then start practice or assessment. Copy the Simulation ID from that screen.

  3. 03

    Call the learner API

    Send Bearer auth to the documented base URL. List your simulations, then address one Simulation ID on every later request.

  4. 04

    Submit assessment

    When the assessment simulation is ready, submit it. Read the safe report, then print the certificate of completion from the result page.

Practice and assessment

Practice explores. Assessment scores.

They are separate simulations. Practice work is never flipped into assessment mode.

Practice
Resettable exploration with check feedback. No certificate.
Assessment
A locked attempt. No reset. One submission produces the safe report and a certificate of completion.
Three cases
The free challenge asks your workflow to handle three refund cases in one assessment simulation. Case answers stay unpublished.

Agent facts

Credentials, IDs, and the API

Every learner operation names a simulation explicitly. There is no implicit active-simulation route.

Sign in
/lab/auth
Dashboard
/lab/dashboard for token creation, practice, assessment, and the Simulation ID shown in the UI.
API base URL
https://api.aiminority.dev
Auth header
Authorization: Bearer <api-token>
Simulation IDs
GET /v1/simulations lists simulations you own with id, purpose (practice or assessment), and lifecycle.
Pinned docs
/lab/docs is the contract, errors, reset, checks, and submit reference.
Agent brief
/lab/agent.md is the same cold-start path as plain markdown for coding agents.

Boundary

Private material stays private

Legitimate operational evidence is available through the documented APIs. Hidden assessment expectations, seeds, and validator internals remain private.

Companies

Need this approach on a real workflow?

The Lab shows how Daniel approaches consequential agentic work. For a live internal system, use the consulting path.

Agentic-engineering consulting