error-recovery

Standard failure recovery patterns for resilient agents include Retry with Backoff, Fallback Alternatives, Diagnose-and-Fix, Escalate with Context, and Graceful Degradation. These patterns help agents adapt and recover instead of merely reporting errors.

Handle transient API errors using the Retry with Backoff pattern, which automatically re-tries transient failures with exponential backoff. This recovery pattern allows agents to adapt when operations fail and service connections degrade.

Implement fallback alternatives by switching to alternative tools or methods when a primary approach fails. This recovery pattern ensures automated workflows continue operating even if a primary service degrades or becomes unavailable.

The best way to escalate unresolved failures is using the Escalate with Context pattern, which escalates unresolved failures with concise, actionable context for human review. This ensures reliability and provides clear logging for agent coordination.

Apply graceful degradation by skipping non-critical steps and documenting the compromise during degraded services. This recovery pattern allows automated workflows to continue delivering value instead of completely failing when operations encounter issues.

Diagnose and fix build or test failures automatically by reading errors, identifying root causes, applying targeted fixes, and re-running. This Diagnose-and-Fix recovery pattern provides clear steps for agent coordination and reliability.