FAQ
Why does the task continue after I close the browser?
AgentHub runs sessions in backend-managed processes. Browser disconnect does not terminate the runtime by design.
Should I use one agent per repository?
Not always. Use one agent per stable intent or workflow. For parallel work, separate agents and worktrees are easier to reason about.
What is the safest default for daily usage?
Use create_worktree mode with predictable naming and clean up stale runs
regularly.
Why do I get path permission errors?
Most often the path is outside safe_paths or resolved path differs from what
you expect. Confirm configured safe path list first.
How do I debug missing output events?
- Check agent state in cards
- Inspect Debug/Raw stream
- Verify server process and logs
- Retry from a fresh session if stream state is unrecoverable
How do I keep long sessions readable?
- Keep prompts narrow
- Split large tasks into phases
- Start a new session when context drifts too far