When a Coding Agent's Task Is Actually Done

A developer argues that an AI coding agent going idle does not mean its task is complete, since the agent may simply be waiting for further input. In the author's view, a task is only finished once the changes have been reviewed, checks have run on the merged result, and the work is on the main branch, so completion is a human judgment rather than an agent signal. The author describes marking tasks done manually when satisfied, at which point they are ready to merge. The piece also notes the opposite problem: without a closing step, waiting tasks accumulate and get forgotten. The author's tool, Styx, auto-archives finished tasks after seven days to keep the list manageable.