My battle to make #llm coding agents follow #tdd continued today. By including the instruction to code like @kentbeck@hachyderm.io I've made some headway. Today the LLM hit a new level:
1. Write a test.
2. Run the test and show me it fails.
...
3. Edit the *test code* so the tests now pass.
4. Declare done with no product code updated.
Sigh.

