# VerifiedX OpenAI Direct support multi-agent starter

Protect a real high-impact support workflow in a composed Python system built directly on OpenAI tool calling.

This starter mirrors a common support topology:

- a triage agent inspects the case and refund controls
- it creates the refund-execution work item for the refund queue
- the refund agent retries the same work item once
- VerifiedX blocks the refund while trust review is still open
- trust review clears the blocker in the case system
- the original refund task is re-dispatched and succeeds

## Run it

```bash
python -m venv .venv
. .venv/bin/activate
pip install -r requirements.txt
verifiedx doctor
```

Set env:

```bash
export OPENAI_API_KEY=...
export VERIFIEDX_API_KEY=...
export VERIFIEDX_BASE_URL=https://api.verifiedx.me
```

Run:

```bash
python run.py
```

What to check:

- `artifacts/scenario_report.json`
- `artifacts/verifiedx_diagnostics.jsonl`

You should see the same refund task stay on the execution queue for one retry, trust review resolve the blocker, and the original refund task succeed on the final dispatch.