Description

When using skills (e.g. mattpocock/skills to-issues) or any scenario that generates long-running task
breakdowns, the TUI exhibits three interrelated issues:

1. Shell PID hang blocks task flow, TUI exits LIVE state prematurely

During long-running tasks that invoke shell for test execution or commands, the spawned shell sub-process may
hang (PID not terminating). This causes:

• The call chain never returns results; the Agent flow stalls waiting for shell output
• TUI exits "LIVE" state due to prolonged inactivity
• TUI notifies the Work list as if the task is done, but the Agent is still waiting

Reproduction:
Any long-running task plan with shell execution steps (e.g. tdd skill, diagnose skill involving
pytest/cargo test).