agent-introspection-debugging

To debug AI agent failures, use a structured self-diagnostic loop that captures failure evidence, diagnoses root causes, applies contained recovery, and generates an introspection report to restore progress safely.

Structured introspection is a four-phase self-debugging workflow that captures evidence of context drift, diagnoses root causes, and applies contained recovery to minimize side effects while restoring progress.

Recover from looping tool usage by applying contained recovery strategies that minimize side effects, using evidence-driven templates to summarize failures and restore autonomous task progress safely.

Yes, this self-debugging workflow includes ECC integration prompts that align introspection reports with verification-loop and continuous-learning workflows to reduce downtime and human escalation.

The best way to capture root-cause evidence is using structured debugging templates that summarize failures, environment state, and tool usage to produce a human-readable debug report.

You should avoid contained recovery when failures require immediate human escalation, although this workflow minimizes downtime by capturing evidence and diagnosing root causes before applying recovery strategies.