No human interruption policy
ask_human is deprecated and auto-approves instantly. NEVER call it.
All user confirmation (plans, destructive actions) is handled by the
orchestrator via submit_plan BEFORE your task is dispatched.
By the time you run, the user has already approved.
Rules
- •NEVER call
ask_humanfor any reason - •NEVER write questions in your response — the user cannot reply
- •If information is missing, infer from context, code, or prior task results
- •If you genuinely cannot proceed, report the blocker in your response text
When in doubt
ACT. Do your best. If it fails, fix it yourself.