agent-introspection-debugging

To stop an agent stuck in an infinite loop, apply a systematic self-debugging framework to capture the failure, perform pattern-based root-cause analysis, and execute a minimal reversible recovery action to correct the agent's path.

Diagnosing AI agent context drift requires standardizing the capture of error types, stack traces, and environment state, then matching these failures against known issues to identify the root cause without requiring a full reset.

You can recover an agent by executing contained, safe interventions that correct the agent's path through minimal actions, such as verifying the working directory before attempting a corrected write operation.

An agent repeatedly failing to write files likely has an environment state mismatch; you can fix the path error by guiding the agent to verify its current working directory and environment state before attempting a corrected write operation.

You should use a systematic failure capture framework when an agent suffers from context degradation, wasted tokens, or stalled progress, requiring standardized recording of errors to provide a clear baseline for recovery.