tao-tdd-pipeline

To enforce test-driven development with an AI agent, you need a pipeline gate that requires red→green→refactor discipline and verifies pytest runs clean before allowing the goal to be marked complete.

To stop an agent from claiming a goal is met before tests are green, apply a conjunctive completion gate that rejects GOAL_MET unless the git diff shows new or modified pytest test files and pytest exits successfully.

A red green refactor loop with repo automation works by first writing failing tests, then implementing code to make them pass, and finally refactoring while a completion gate verifies pytest output remains green.

Yes, the pytest completion gate works with existing test conventions by requiring the worker to modify test files matching pytest conventions and successfully run python -m pytest -q tests/.

Your agent workflow returns a discipline violation during TDD when the loop result conflicts with the TDD gate, specifically when test modifications are missing in the git diff or pytest results are not green.