stuck-recovery

You detect repetitive failure loops by monitoring tool calls and file modification patterns using sliding window pattern matching on process-scoped history. This identifies stuck states and triggers advisory outputs for recovery.

Recovery strategies for stuck AI agents include context resets to clear confusion, model escalation, or human intervention requests. These actionable advisories trigger automatically when repetitive failure patterns are detected in the orchestration layer.

To stop an AI agent from repeatedly failing on the same file edit, enable stuck state monitoring to detect the repetitive failure loop. The system then advises a fresh context reset to clear the agent's confusion and resume progress.

Loop detection operates within the orchestration layer by monitoring process-scoped history. It leverages sliding window pattern matching to identify stuck states during active sessions and provide recovery advice without requiring external intervention.

Sliding window pattern matching for error handling relies on process-scoped history, meaning it detects repetitive failures based on monitored tool calls and file modifications. It provides advisory outputs for recovery but does not autonomously execute the fixes.