long-running-taskslisted

# /long-running-tasks Choose the right mechanism when a task will take more than a few minutes. ## Decision tree ``` Task duration estimate? ├─ < 30 seconds → run synchronous (default) ├─ 30s – 2 min → run synchronous, watch for timeout ├─ 2 – 10 min → Bash with `run_in_background: true`, then Monitor for completion ├─ 10 min – 1h → background subagent (`Task` with `run_in_background: true`); main agent continues with other work └─ > 1 hour → consider refactoring; or use a CronCreate for scheduled recurrence; consider whether the model session itself will reach context limits before completion ``` ## Tool reference ### `Bash` with `run_in_background: true` For shell commands > 2 min: ```python Bash(command="<long-running-cmd>", run_in_background=True, timeout=600000) ``` Returns control immediately. Use `Monitor` (or `BashOutput`) to check status. ### `Monitor` tool Watches a background process for events. **Costs no tokens while waiting** (event-driven, not polled): ```python Monitor(...) ``` Wakes the agent only when an event of interest occurs. Use for "wait for build to finish" / "watch a log file for an error pattern". ### Background subagent (`Task` with `run_in_background: true`) For tasks needing reasoning + tool use, not just shell: ```python Task( description="<one-line>", subagent_type="general-purpose", prompt="<self-contained prompt — the agent has no context from this conversation>", run_in_background=True, ) ``` Subagent runs ind