olam-dispatch

Dispatch to Olam World

Send a task prompt to a running world's Claude Code agent.

Workflow

1. Parse arguments: first token is the world ID, rest is the prompt.

2. If no world ID provided, call olam_list to show available worlds and ask which to dispatch to.

3. If no prompt provided, ask the user what task to dispatch.

4. Call olam_dispatch with:

   • worldId — the world ID
   • prompt — the task description

5. Present results showing:

   • Session ID
   • Cost
   • Number of turns
   • Result summary
   • Duration

Cloudflare mode

For CF-hosted worlds, dispatch requires in-container auth to have completed first (OAuth Claude or device-code OpenAI, depending on agent). If the caller hits a 401 from the container's /dispatch, the operator must open $OLAM_WORKER_URL/sandbox/<id>/ and complete auth, then retry.

The task field on olam_create triggers auto-dispatch after auth completes, which avoids this roundtrip on world creation. Reference: docs/CF_WORLDS_SPEC.md §4.3 + §5.

Failure handling

• 401 from container /dispatch (CF mode) — in-container auth incomplete. Tell the operator: open $OLAM_WORKER_URL/sandbox/<id>/, complete the OAuth flow, retry the dispatch.
• olam_dispatch returns "world not running" — call olam_status to confirm state. The world may have been destroyed or its container exited.
• Dispatch returns instantly with no result — the agent inside the world may not be hooked up. Verify via olam_status and the world's dashboard.
• High cost value or turn count — surface this to the operator so they can interrupt early via olam_destroy if the world is in a loop.

Anti-scope (what this skill is NOT for)

• Creating a new world — see olam-create. Dispatch requires an existing running world.
• Watching the agent's reasoning live — see olam-observe (separate verb; dispatch returns when the agent finishes a turn, observe streams the trace).
• Saving thoughts after a successful dispatch — see olam-crystallize.
• Re-running the same prompt — dispatch is idempotency-less; the agent treats every call as a new task.

See also

• olam-create — for new worlds.
• olam-list — find a world to dispatch to.
• olam-status — confirm a world is dispatch-ready.
• olam-crystallize — save the agent's thought graph after dispatch.