Auto-Recovery, Agent Lifecycle & Workspace Policies

Agents now automatically recover from transient failures instead of crashing. A new lifecycle system tracks agent state from start to finish, and workspace policies let you control agent behavior with declarative rules.

• ›Auto-recovery -- agents now automatically retry when they hit temporary issues like service overloads or rate limits, with smart backoff and escalation
• ›Agent lifecycle tracking -- every agent task now has a clear state (starting, running, paused, completed, failed) for better observability
• ›Workspace policies -- set rules to control agent behavior: model selection, budget caps, tool restrictions, and event routing, all evaluated instantly
• ›Recovery notifications -- see real-time updates when an agent is recovering from an issue, so you know what's happening behind the scenes
• ›Improved error messages -- clearer explanations when things go wrong, with context about what the system tried to do about it