Test-First Building

The Builder agent follows strict test-driven development: it reads the specification, writes failing tests for each acceptance criterion, then implements the code to make those tests pass. This ensures 100% of the specified requirements have corresponding test coverage, and the implementation is verified against the original design.

How it works

The Builder agent reads the specification and creates test files first. Each acceptance criterion from the spec becomes at least one test case. Tests are run to confirm they fail (proving they actually test something). Then the agent implements the feature, running tests iteratively until all pass. If a test reveals a spec ambiguity, the agent flags it for review rather than guessing. The test suite becomes a regression safety net for future changes.

Terminal showing test-first workflow with failing then passing tests

Screenshot coming soon

Why it matters

AI coding without tests is a recipe for regressions. The AI might produce code that works for the happy path but breaks edge cases or existing functionality. TDD flips this: the AI must prove its code works against explicit criteria before moving on. This approach also makes code review much easier -- reviewers can check test coverage against the spec instead of manually reasoning about correctness.

Test coverage report showing spec criteria mapped to test cases

Screenshot coming soon