agent-introspection-debugging

What problem does it solve? AI agents often fail silently by looping on the same tool calls, burning tokens without progress, or drifting from the original task. This Skill provides a structured self-debugging workflow so the agent captures the failure, diagnoses the root cause, applies a contained recovery, and reports the outcome instead of retrying blindly. ## Core Features & Use Cases - Four-Phase Loop: Failure Capture, Root-Cause Diagnosis, Contained Recovery, and Introspection Report, each with ready-to-use Markdown templates. - Failure Pattern Table: Maps common symptoms (max tool calls, context overflow, ECONNREFUSED, 429 quota exhaustion, stale file state) to likely causes and verification checks. - Recovery Heuristics: Ordered interventions from restating the objective to running one discriminating check before any retry. - Use Case: An agent hits its tool-call limit while repeatedly running the same failing test command. The skill guides it to capture the error, classify it as a loop pattern, verify actual filesystem state, narrow scope to one failing test, and produce a structured debug report for the human. ## Quick Start Use the agent-introspection-debugging skill to diagnose why the current agent run is stuck in a loop and produce a structured recovery report.