How to setup coding orchestration: the agent finishes one task, then stops
A developer hands a backlog to a coding agent, but after each task the agent reports a blocker and refuses to pick up the next one, even though it has all the access it needs. The fix the thread points to is structural, not a better prompt: define an explicit done-check per task, keep the backlog in a file the agent updates itself, and make it print exactly what it tried when it claims to be blocked.