TDD for Agent Loops

Inspired by mattpocock/skills tdd — adapted for LangGraph.

Red → Green → Refactor

1. Red — Write failing test for routing or node output (mock LLM)
2. Green — Minimal node/routing fix
3. Refactor — Extract helpers only when second test needs it

What to test

Fixtures

@pytest.fixture
def fake_llm():
    return FakeLLM(responses=[{"content": "..."}])

Async

@pytest.mark.asyncio
async def test_graph_happy_path(tmp_path):
    ...

Avoid

• Tests that only assert "graph compiles"
• Snapshotting full LLM prose (flaky)
• Hitting live Groq in CI

Reference

• loop-engine-agent-platform/tests/test_repo_fix.py
• *-agent-pattern/tests/ — minimal pattern coverage