agent-introspection-debugging

What problem does it solve?

This Skill helps capture and diagnose agent runs that loop, consume excessive tokens, or fail repeatedly by guiding a systematic four-phase self-debugging workflow: failure capture, root-cause diagnosis, contained recovery, and an introspection report.

Core Features & Use Cases

• Failure Capture: Record error details, last meaningful tool calls, environment assumptions, and context pressure before retrying.
• Root-Cause Diagnosis: Classify failures into common patterns (looping tool calls, context overflow, network/quota errors, stale filesystem state) and derive a minimal hypothesis.
• Contained Recovery: Apply the smallest safe corrective action such as trimming context, isolating a failing command, re-checking environment state, or stopping retry storms.
• Introspection Reporting: Produce a structured, human-readable debug report that documents the failure, diagnosis, recovery action, evidence, and follow-up.
• Use Case: Recover an autonomous agent stuck in repeated tool calls or drifting from its goal by capturing state, running a discriminating check, and producing a clear handoff for human review.

Quick Start

Run the agent-introspection-debugging workflow to capture the failure state, diagnose the root cause, perform a minimal contained recovery, and produce an introspection report.