When a parallel agent gets stuck, the workflow should pause that task, surface the blocker, and let the user decide whether to clarify, retry, change models, or stop the task. A stuck agent should not keep rewriting code blindly.
Common blockers include unclear requirements, failing tests with no obvious cause, missing credentials, unavailable services, dependency conflicts, or overlapping changes from another agent. The fix depends on the blocker. Sometimes the right answer is more context. Sometimes it is a smaller task. Sometimes it is human review.
Verdent's useful role is visibility and control. In a parallel workflow, one stuck worker should not stop all work. Other agents can continue in their own isolated workspaces while the blocked task waits for a decision. That is why a good agent deck matters: it shows which work is moving, which work needs attention, and what should be reviewed next.
