| name | optimus-fleet |
| description | TEMPORARILY DISABLED. Do not use this skill for automatic Optimus fleet dispatch. The Optimus orchestrator passthrough is currently disabled because wait_for_completion can exceed MCP request timeouts; when users ask for optimus-fleet, fleet mode, or automatic orchestration, explain that the skill is disabled and use manual step-by-step execution or direct sub-agents instead. |
| license | MIT |
Optimus Fleet Orchestrator — Temporarily Disabled
This skill is intentionally disabled until the underlying Optimus MCP orchestration timeout is fixed and validated.
Why this is disabled
The previous implementation forced a strict passthrough call to optimus_orchestrate with wait_for_completion=true and a long completion timeout. In current MCP clients, long-running tool calls can exceed the MCP request timeout and fail with:
McpError: MCP error -32001: Request timed out
That failure mode is worse than a normal task failure because it makes the orchestration entrypoint itself unreliable and can leave the user believing work is being managed when the controlling request has already timed out.
Required behavior while disabled
When this skill triggers:
- Do not call
optimus_orchestrate as a passthrough.
- Tell the user that
optimus-fleet is temporarily disabled because the orchestrator can hit MCP request timeouts.
- Continue with a safer manual workflow:
- decompose the task into explicit todos,
- dispatch direct sub-agents with the
task tool when useful,
- use local tools for implementation and verification,
- keep the user informed that the result is not fleet-native end-to-end orchestration.
- If the user explicitly asks to re-enable fleet, explain that the source fix should first cap or redesign
wait_for_completion so MCP calls return before the client request timeout, then add regression tests for the timeout path.
Re-enable criteria
Only restore strict fleet passthrough after all of the following are true:
optimus_orchestrate, delegate_task_async, and dispatch_plan_async no longer block an MCP request longer than the MCP client timeout.wait_for_completion=true either uses an MCP-safe cap or is replaced by an async polling workflow.- Regression tests cover timeout capping, default non-blocking behavior, terminal status polling, and failure reporting.
- A real end-to-end fleet run completes or returns a controlled non-timeout result.
