Usage:
  kent run [flags] <prompt>
  kent run steer [--persistence-root <root>] <session-id> <message...>
  kent run stop [--persistence-root <root>] <session-id>
  kent run wait [--persistence-root <root>] [--output-mode=json] <session-id>
  kent run watch [--persistence-root <root>] [--output-mode=json] <session-id>

Run a headless subagent session. This will start a real Kent agent with optional specified role and a specific prompt to stay quieter during work.

- Headless subagents cannot ask questions and will try to work until they hit a blocker or do as much as they can.
- For this command to work, a shared Kent server/service must be running (see relevant documentation).
- Finalized assistant responses print as they are committed. Use `-q` or `--quiet` for final-result-only output.
- New sessions print a ready-to-use `run steer` command after the run becomes steerable.
- Prefer passing `--session <id>` (returned by the first run of the `run` command) to continue the previous session.
- Prefer continuing sessions over starting new ones for re-reviews, continuation of work, follow-ups, discussion, and start new ones separate tasks.
- If you're an agent, you can't control your own session.

Active-run controls:
  `steer` queues a message for an active run.
  `stop` interrupts an active run.
  `wait` waits for an active run to finish.
  `watch` observes the next outcome of an active run.

Subagents:
  `--agent <role>` selects a named role from `[subagents.<role>]` section of the user configuration. For agents, available agents are the default (no flag), fast, and any listed in the developer reminder you received.
  `--fast` is shorthand for the built-in `fast` role.

Examples:
  kent run "summarize the unstaged changes"
  kent run --session <session-id> "follow-up"
  kent run steer <session-id> "use the safer migration path"
  kent run wait --output-mode=json <session-id>
  kent run watch --output-mode=json <session-id>
  kent run --fast "scan the repo"

Flags:
