Examples¶

Customer Service Agent¶

A complete example demonstrating understudy with a customer service agent is available in the example/ directory.

The example includes:

A customer service agent with order lookup and return handling
Scene files for various scenarios (returns, policy enforcement, adversarial)
Test files showing deterministic assertions and LLM judge usage

Running the Example¶

Install dependencies:

pip install understudy[all]
# or
uv add understudy[all]

Set your API key:

export ANTHROPIC_API_KEY=your-key
# or
export GOOGLE_API_KEY=your-key

Run the simulation:
```
cd example
python run_simulation.py
```
Run tests:
```
pytest example/test_returns.py -v
```

Scene Files¶

The example includes these scenes:

return_eligible_backpack.yaml: Customer returns a backpack (eligible category). Agent should process the return.
return_nonreturnable_earbuds.yaml: Customer returns earbuds (non-returnable category). Agent should deny.
adversarial_policy_bypass.yaml: Customer attempts social engineering to bypass return policy.

CI Integration¶

Example GitHub Actions workflow:

name: Agent Tests
on: [push, pull_request]

jobs:
  test:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-python@v5
        with:
          python-version: "3.12"
      - run: pip install understudy[all] pytest
      - run: pytest tests/ --junitxml=results.xml
        env:
          ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
      - uses: dorny/test-reporter@v2
        if: always()
        with:
          name: Understudy Results
          path: results.xml
          reporter: java-junit