deliberation-tester

Test-driven development for adapters starts by writing a failing unit test that expects your adapter's `parse_output` method to correctly extract the model response from raw CLI output. Run `pytest tests/unit/test_new_feature.py -v` to verify the test fails, then implement the adapter to make it pass, ensuring robust parsing logic from the start.

Use VCR cassettes with vcrpy to record and replay HTTP responses. This approach ensures consistent, efficient integration tests for HTTP adapters by capturing real responses once and replaying them in subsequent test runs, eliminating flaky network dependencies.

Use pytest-asyncio to write and run async unit tests for your deliberation logic. This dependency enables the red-green-refactor cycle for async AI features, allowing you to test concurrent adapters and convergence logic reliably without blocking.

Yes. TDD ensures your CLI adapter correctly extracts and parses model responses. Write a unit test first that validates `parse_output` behavior on sample CLI output, then implement the adapter to pass that test, reducing parsing bugs and improving reliability.

Apply TDD with mocked adapters and subprocesses for fast, isolated unit tests of convergence logic. Mock external dependencies to test how your deliberation feature handles adapter outputs, then use VCR cassettes for integration tests that verify end-to-end behavior without live HTTP calls.

Mocking adapters and subprocesses isolates your tests from external dependencies, making them fast and deterministic. This allows you to test edge cases and failure scenarios reliably, while VCR cassettes handle HTTP responses for integration tests without introducing network brittleness.