Spawns real AI-powered OpenClaw sub-sessions to run multiple specialized agents concurrently for content, dev, QA, docs, and autonomous workflows.
Smart model hierarchy: Haiku → Kimi → Opus. Auto-escalation for cost optimization (~57% savings). Agents now try cheapest model first, escalate to more powerful models if needed.