| name | boomerang |
| description | Execute a task autonomously in an isolated forked context and return only a compact summary, keeping the main conversation free of the work's token weight. Use for self-contained tasks whose intermediate steps don't matter to the ongoing conversation. |
| argument-hint | <task> |
| context | fork |
| agent | general-purpose |
Execute this task autonomously:
$ARGUMENTS
Rules:
- Work without asking questions. Make reasonable assumptions and note them in
the summary instead of stopping.
- Follow the global instructions in CLAUDE.md (jj hygiene, quality gates),
except: do NOT commit — leave that to the main conversation.
- Stay within the task's scope; no opportunistic refactors.
Your final message is the only thing the main conversation will see. Make it
this compact summary and nothing else:
## Boomerang: <task in a few words>
**Outcome:** done | partial | blocked — one sentence.
**Changes:** file paths with one-line notes (or "none").
**Commands run:** the ones worth re-running for verification.
**Assumptions:** any decisions made without asking.
**Follow-ups:** what the main agent or user still needs to do (or "none").