What problem does it solve? AI agents often stall mid-task by asking questions in plain prose, which ends the turn without completing the work. This Skill enforces a consistent rule: every decision, clarification, or approval request goes through the structured question tool, and when no user is available, the agent resolves the decision itself and keeps going. ## Core Features & Use Cases - Structured Questioning: Presents decisions as labeled options or open-ended free-text prompts via the question tool, one question per concern. - Autonomous Fallback: Handles two no-user scenarios — the question tool being absent, or a [Never-Ask] response — by re-evaluating options for headless execution and proceeding. - Safety Guardrails: Never auto-approves destructive, irreversible actions like deleting branches or dropping data; prefers text-only, non-interactive, minimal-scope paths. - Use Case: During an unattended eval run, the agent needs to pick an auth strategy. Instead of stopping to ask, it selects the best non-interactive option, states its reasoning, and continues the task. ## Quick Start Ask the agent to use the compose:ask skill whenever it needs a decision or approval so it routes questions through the question tool and never stalls the task.
