Human in the loop approval

**What the feature is (as Title)**
Gate any sensitive tool behind explicit human approval. The agent pauses, surfaces the proposed action and its reasoning in a structured form, and waits for approval or rejection, with every decision captured in a full audit trail.

**Value proposition**
What: Mark any sensitive tool to require approval, and the agent pauses before running it, showing the proposed action and its reasoning so you can approve or reject. The approval form is schema-driven and consistent across Agent Builder and Workflows.

Who: Developers who want the agent to act autonomously, with manual approval required on sensitive actions.

Why it matters: Developers stay in control of sensitive actions, and every approval or rejection is recorded in a full audit trail.