Agent Introspection Debugging
Implements a structured self-debugging workflow to help AI agents recover from loops, errors, and task deviations.
The Agent Introspection Debugging skill provides a systematic four-phase framework—Capture, Diagnosis, Containment Recovery, and Introspection Reporting—to prevent AI agents from wasting tokens on repetitive failures. By moving away from blind retrying, it enables agents to analyze the root cause of issues such as tool loops, context drift, and environment mismatches. This skill ensures that when an agent hits a wall, it pauses to diagnose the state, prunes unnecessary context, and generates a human-readable report before escalating or attempting a more informed recovery.
주요 기능
01Automated context pruning to maintain reasoning quality and reduce token waste
02Pattern matching for common AI failures like tool loops and context overflow
03Integration with ECC ecosystem for continuous learning and verification
04Detailed introspection reports for transparent human-in-the-loop auditing
05214,597 GitHub stars
06Four-phase recovery loop: Capture, Diagnose, Recover, and Report
사용 사례
01Recovering from reasoning degradation caused by context window saturation and prompt drift
02Diagnosing environmental state mismatches such as port conflicts or file system drift
03Breaking infinite loops where an agent calls the same tool repeatedly without progress