@openma/cli@0.5.0

Minor Changes

• #103 85c2e1c Thanks @hrhrng! - oma bridge daemon now serves multiple tenants from a single process.
  One daemon is authorized for every workspace the user is a member of;
  each spawned ACP child gets the per-tenant oma_* key matching the
  session's workspace.

  • oma bridge setup requests the multi-tenant /exchange shape and
    writes a CredentialsV2 file ({v:2, tenants:[…], …}). Old v1
    creds files (agentApiKey at the top level) auto-migrate on next
    daemon start — calls GET /agents/runtime/me to pull the tenant
    list, falls back to a placeholder workspace if the server is
    unreachable so the daemon still runs offline.
  • oma bridge refresh (new) re-syncs the daemon's credentials with
    the user's current memberships. Adds keys for new workspaces, soft-
    revokes keys for removed ones, then SIGHUPs the running daemon
    so the change takes effect without a restart.
  • SessionManager looks up the right oma_* key per session by the
    inbound session.start's tenant_id. Every outbound message the
    daemon sends carries tenant_id so the server can validate it
    against the runtime's authorized set.

  Backward-compatible: v1 daemons keep working against the new server
  shape (server returns the legacy {runtime_id, token, agent_api_key}
  when the request doesn't set multi_tenant: true). The workaround
  for multi-tenant — running multiple OMA_PROFILE=… daemons side by
  side — still works for separate server environments.

Patch Changes

• 31f7fbf Thanks @hrhrng! - oma bridge setup now exits cleanly after "Done." instead of hanging
  for ~5 minutes on idle keep-alive HTTP sockets from the registry CDN
  fetch and the runtime-token probe. Daemon was already started by
  launchd / systemd / Task Scheduler — only the foreground setup process
  itself was waiting on the undici dispatcher to time out its sockets.
  Force-exits at end of runSetup, matching how npm / pnpm / gh handle
  the same constraint in their CLI commands.

  Adds an opt-in OMA_DEBUG_HANDLES=1 env var that prints active
  handles + requests every 2s — useful for diagnosing future "process
  won't exit" regressions without redeploying.