This file is the prompt sent to the agent on every iteration. Replace its contents with your own task. Because each iteration starts a fresh session, write the prompt so the agent rebuilds context from the workspace every time.
A good loop prompt typically tells the agent to:
- Read the current plan / notes / TODO in this repository.
- Make the smallest useful increment of progress toward the goal.
- Run the tests and record what changed.
- Create a
stop.mdfile when the work is complete so the loop exits cleanly.