Backend & API Testing

Backend tests in XAIO typically run as pytest in the workspace container. The agent sets up a tests/ structure if not already present, and can write unit, integration, and smoke tests.

Endpoint tests

Test all endpoints at once:

> Test all FastAPI endpoints for me. For each: 200, 400, 401, 404 where applicable. Write it in tests/test_endpoints.py and run it.

What happens: The agent reads your routes from main.py, generates test cases per route with valid JWT, invalid payload, missing auth header — and runs them via pytest -v.

Per-endpoint detail:

> Write a test for POST /api/projects — it must verify that:

> 1. A logged-in user can create a project

> 2. Plan-limit violation returns 402

> 3. Invalid pipeline_type returns 400

Database tests

Migration testing:

> Test migration V075. Apply it, create some beta applications, verify all fields persist correctly. Then roll it back and verify the table is gone.

Constraint tests:

> Write a test that verifies status can only be 'pending', 'approved', 'rejected', 'waitlist', or 'contacted'. Other values must throw an IntegrityError.

Mock vs. real DB

By default, the agent uses a real local Postgres for backend tests (see memory entry *integration tests must hit a real database, not mocks*).

If you explicitly want mocks:

> Mock the DB calls for this test — I want to test the service logic in isolation, not persistence.

External APIs (Stripe, Brevo, GitHub)

These are always mocked, never called for real during tests:

> Test the Stripe webhook handler. Mock a payment_intent.succeeded event, verify that the subscription is set to active, idempotency test with a duplicate event ID.

Property-based testing

For complex invariants, the agent can use hypothesis:

> Use hypothesis to verify that the slug generator is ALWAYS URL-safe, no matter what input. Try Unicode, emojis, very long strings.

Integration tests with containers

For tests that need a real Postgres + Redis:

> Write an integration test that creates a user, creates a project, publishes it, and verifies all 6 cleanup steps work when deleting it.

Coverage & CI

> Run pytest --cov=services --cov-report=html and tell me where we're under 80% coverage.

The agent can also generate a GitHub Actions workflow file:

> Write a .github/workflows/test.yml that runs the backend tests on every PR — Postgres as a service container.