What problem does it solve? AI agent runs often fail by looping on the same tool calls, burning tokens without progress, or drifting from the original task, and blind retries make things worse. This Skill gives the agent a disciplined four-phase workflow to debug itself before escalating to a human. ## Core Features & Use Cases - Failure Capture: Records the error, last tool calls, context pressure, and environment assumptions before any retry. - Root-Cause Diagnosis: Matches failures against a pattern table covering tool-call loops, context overflow, connection errors, rate limits, and stale filesystem state. - Contained Recovery: Applies the smallest reversible corrective action, such as trimming context, verifying world state, or narrowing scope to one failing test. - Introspection Report: Produces a structured self-debug report with root cause, recovery action, result, and preventive follow-ups. - Use Case: An agent hits its maximum tool-call limit while retrying the same failing command. The skill captures the failure, classifies it as a loop pattern, verifies actual filesystem state, and produces a report explaining the root cause and fix. ## Quick Start Ask the agent to run the introspection debugging workflow on the current failing session and produce a self-debug report.
