What problem does it solve? In automated agent workflows, human approval steps can be silently skipped or auto-completed, breaking accountability. This Skill defines the human actor's task spec in the open-autonomy hello-human example so approval requests genuinely reach a person and wait for explicit confirmation. ## Core Features & Use Cases - Human-in-the-loop handoff: The requester agent engages the approver through the runner seam, which parks the ask and never completes it on its own. - Multiple notification channels: Asks are printed to the console, appended to .open-autonomy/runner-state/human-attention.md, and optionally sent via a configured AUTONOMY_HUMAN_ENGAGE_CMD hook. - Verified completion: The flow only resumes when the person runs bun scripts/runner.ts update <id> --status done, guaranteeing completion is verified rather than presumed. - Use Case: In a local open-autonomy installation, a requester agent needs sign-off before proceeding; the approver spec ensures the request reaches a human and blocks until that human explicitly marks it done. ## Quick Start Watch the human-attention file for new approval asks, then mark the parked session done with the runner update command once you have reviewed it.
