check Todo list for TDD approach

[MPietrala]

The hook is currently checking the todo list before executing and saving the output, but it is not reinforcing TDD in the to do list. If we would already enforce it on this level, then we would save a lot of back and forth with claude.

[MPietrala]

Also just want to say. Amazing little tool, it makes claude code a bit slower, but the output is so much cleaner and things just work in one go. Thanks!!!

[nizos]

Excellent point, this is definitely an area worth exploring!
Thanks so much for the kind words and the thoughtful input!! :)

[nizos]

Hi again!

I have started experimenting with this idea and here is what I found:

• Refactoring is much more meaningful, especially in tests
• The agent does not try to write a full implementation in one go before being corrected
• It still tries to over-implement sometimes, but not as egregiously

To my knowledge, we can't modify the agent's todos directly, but we can prompt it to make the changes we want. I had it explicitly state "following strict TDD principles" for development tasks and added explicit refactoring steps.

"todos": [
  {
    "content": "Implement cart add operation following strict TDD principles",
    "status": "pending",
    "priority": "high",
    "id": "1"
  },
  {
    "content": "Refactor cart and tests after implementing add functionality",
    "status": "pending",
    "priority": "high",
    "id": "2"
  },
  {
    "content": "Implement cart remove operation following strict TDD principles",
    "status": "pending",
    "priority": "high",
    "id": "3"
  },
  {
    "content": "Refactor cart and tests after implementing remove functionality",
    "status": "pending",
    "priority": "high",
    "id": "4"
  }
]

I see two approaches:
Option 1: Developers create a custom sub-agent for planning that ensures this structure

• Pros: Maximum control and flexibility
• Cons: Requires manual configuration

Option 2: We implement opt-in functionality that automatically guides the agent when it modifies todos

• Pros: Works automatically and seamlessly
• Cons: Might interfere with existing todo management systems/subagents

Option 2 is straightforward to implement: We already listen to todoWrite operations and can respond with guidance to ensure desired structure.

The challenges:

• Preventing todo items from being dropped (length limits)
• Avoiding confusion when todos are already compliant
• Preventing instruction fatigue (agent ignoring repeated prompt)
• Avoiding infinite loops after compliance
• keeping non-dev tasks unaffected (docs, publishing, investigation)

Potential solution: Since we already store todos in todos.json, we could validate todo compliance using the same validation process we use for file modifications.

What are your thoughts on this approach? :)

[MPietrala]

I would personally go for option 2. I was also thinking that it would have been the similar validation process as with the other hooks. And those sub agents become again another thing to maintain. Think the beauty of this tool is, that it is super easily to implement.