SuperAPI AI Directory

alera-orchestration

Agents & models

View repository

Updated 10-15 23:59

About this skill

Alera Orchestration

The runtime host owns lifecycle authority. Use its lifecycle commands, not generic messages or arbitrary status edits, for completion, cancellation, leases, and ownership.

Choose Your Role

  • Dispatched worker: accept first with alera orchestration dispatch-accept, then read alera orchestration --json context and worker. Follow the installed task and worker instructions.
  • Coordinator dispatching work or managing a run: read coordinator.
  • Stalled tasks, cancellation, gates, or ownership transfer: read recovery.
  • Message delivery or terminal inspection: read messaging.

Read only the references relevant to the current role. Worker completion is atomic and required exactly once; after successful completion, stop the turn and leave the terminal available for reuse.

Shared Constraints

Run inside an Alera terminal. ALERA_TERMINAL_HANDLE identifies the caller and ALERA_WORKSPACE_ID supplies the workspace default; alera orchestration current inspects both. Check alera version --json if compatibility is uncertain.

Scope every task to its workspace and coordinator. A workspace may have one active coordinator run; run tasks and injected workers must match that run's workspace and coordinator.

Honor prior explicit user authorization for the same scope without duplicate approval requests. Runtime policy approvals and user-owned stall gates still apply; never bypass them or invent a user decision. Active workers ask the coordinator through alera orchestration ask, not a local user-input prompt.