Summary

The spec defines a Checkpoint Recovery strategy (§6.6 Strategy 2) that persists AgentContext snapshots after each completed turn and resumes from the last checkpoint on crash. Only Fail-and-Reassign is implemented.

Design Spec Reference

• §6.6 Agent Crash Recovery — Strategy 2: Checkpoint Recovery

Scope

• Per-turn AgentContext checkpoint persistence (leverages model_dump_json())
• Heartbeat-based failure detection
• Resume from last checkpoint with environment reconciliation
• Max resume attempts before falling back to fail-and-reassign
• RecoveryStrategy protocol compliance