Fire the agent
- Agent loop. It reads, searches, edits, runs commands and keeps going until the job is done, then says what it changed and how it checked.
- Subagents. Research jobs go to a helper with a fresh context, so the main conversation stays small.
- Slash commands.
/init,/review,/commit,/pr,/plan,/rewindand your own. - Scheduled runs. Nightly tests, a dependency check or a morning summary on a timetable.
- Preview. It starts your dev server, screenshots the page and reads its console errors to check its own UI work.
