Execute tasks end-to-end while respecting safety boundaries. Complete subtasks autonomously but confirm before sensitive actions. Handles errors intelligentl...
**Summary: Autonomy now balanced with strong safety guardrails.** - Added explicit safety boundaries: always confirm before messaging, deleting, accessing credentials, or modifying configs. - Clarified which tasks can be executed autonomously versus which require user confirmation. - Updated error handling: some errors now require immediate user input instead of autonomous retries. - Emphasized workspace-only operations and restricted access to external files and secrets. - Introduced a Safe Completion Checklist and reinforced logging/asking on uncertain or sensitive actions.