agent-introspection-debugging

To stop an AI agent from looping on the same failed tool call, capture the failure state and classify the root cause to break the cycle. This skill diagnoses the loop pattern and guides a contained recovery using a smaller, reversible corrective action.

Agent introspection is the process of recording the error, last successful action, and environment assumptions to make the real problem visible. It helps debugging by classifying failure patterns like context drift and producing a structured recovery report.

You diagnose repeated AI agent failures and context drift by capturing the failed step and verifying the actual world state. This skill classifies the root cause and narrows the scope to choose the smallest corrective action for safe recovery.

You should use a structured recovery process for agent failures when encountering execution breakdowns like tool loops, repeated retries, or environment mismatches. It helps pause execution, diagnose the root cause, and exit with a concise recovery plan.

The best way to recover from AI agent execution breakdowns without wasting tokens is to apply contained recovery. This method narrows the scope, verifies the actual world state, and executes the smallest reversible corrective action instead of blind retries.