What problem does it solve?
It helps an AI agent stop wasting tokens on repeated failures by capturing the exact failure state, diagnosing the most likely agent-specific pattern, applying a small contained recovery, and producing a structured self-debugging report for the next run or a human reviewer.
Core Features & Use Cases
- Failure capture: records the error, last meaningful tool sequence, context pressure, and environment assumptions before attempting any recovery.
- Root-cause diagnosis: classifies common failure patterns such as tool loops, context overflow, service connectivity issues, quota exhaustion, and filesystem/cwd drift.
- Contained recovery + reporting: performs the smallest reversible action, then emits a clear introspection report including what changed and what evidence indicates improvement.
- Use case example: when an agent repeatedly calls the same tools, loops on a command, or drifts due to prompt/context growth, this workflow produces a human-readable diagnosis and next-step plan rather than continuing retries.
Quick Start
Use the agent-introspection-debugging workflow when your agent is failing repeatedly or looping, and ask it to produce a full self-debug report using the four phases: failure capture, root-cause diagnosis, contained recovery, and an introspection report.
