Context

When an agent execution fails unexpectedly (unhandled exception, OOM, process kill), the framework needs a recovery mechanism. The MVP implements the fail-and-reassign strategy behind a RecoveryStrategy protocol, enabling future addition of checkpoint-based recovery without modifying existing code.

Acceptance Criteria

RecoveryStrategy Protocol

• RecoveryStrategy protocol defined with recover(task, error, context) method
• Protocol is pluggable — new strategies can be registered via config

Fail-and-Reassign Strategy (Default / MVP)

• Engine catches failures at its outermost boundary
• Redacted AgentContext snapshot logged on failure (turn count, accumulated cost — no message contents)
• Task transitions to FAILED status
• Failed tasks are available for reassignment (manual or automatic via task router)
• max_retries configurable per task (default: 1)
• Reassignment respects retry count — tasks exceeding max retries stay FAILED

New TaskStatus: FAILED

• FAILED added to TaskStatus enum as a non-terminal state
• Valid transitions updated: IN_PROGRESS → FAILED, FAILED → ASSIGNED (reassignment)
• FAILED differs from CANCELLED (terminal) — failed tasks are eligible for reassignment

Testing

• Unit tests for RecoveryStrategy protocol
• Unit tests for fail-and-reassign with mock agent failures
• Integration test: agent fails → task marked FAILED → reassignment available

Dependencies

• Implement agent engine core with ExecutionLoop protocol integration (DESIGN_SPEC §3.1, §6.1, §6.5) #11 — Agent engine core
• Implement single-task execution lifecycle (assign, execute, complete) #21 — Task execution lifecycle (status transitions)

Design Spec Reference

• §6.6 — Agent Crash Recovery (Strategy 1: Fail-and-Reassign)