Verified:GitHub

agent-introspection-debugging

Diagnoses and recovers from AI agent failures through structured capture, root-cause analysis, and introspection reports.

Updated Jul 6, 2026

Search Tags:#root cause analysis #ai agent #introspection #agent debugging #failure diagnosis #self-debugging #recovery workflow

Ready to run

npx skills add https://github.com/chenziyang110/launchdeck --skill agent-introspection-debugging-chenziyang110

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: agent-introspection-debugging
Source: https://github.com/chenziyang110/launchdeck/tree/main/.claude/skills/agent-introspection-debugging
Command: npx skills add https://github.com/chenziyang110/launchdeck --skill agent-introspection-debugging-chenziyang110

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? AI agents often fail silently by looping on the same tools, burning tokens without progress, or drifting from the original task. This Skill provides a structured self-debugging workflow so the agent captures failure state, diagnoses root causes, applies contained recovery, and reports results instead of blindly retrying. ## Core Features & Use Cases - Four-Phase Loop: Failure capture, root-cause diagnosis, contained recovery, and a structured introspection report. - Failure Pattern Library: Maps common symptoms (max tool calls, context overflow, ECONNREFUSED, 429 quota exhaustion, stale file state) to likely causes and checks. - Recovery Heuristics: Prioritizes restating the objective, verifying world state, shrinking scope, and running one discriminating check before retrying. - Use Case: When an agent hits a loop-limit failure while fixing a test, activate this Skill to capture the failure, classify the pattern, run one direct check, and produce a human-readable self-debug report. ## Quick Start Ask the agent to use the agent-introspection-debugging skill to diagnose why its current task is stuck in a retry loop and produce a self-debug report.

Frequently Asked Questions about agent-introspection-debugging

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I debug an AI agent stuck in a tool-call loop?▼

Capture the failure state first: error message, last tool calls, and current goal. Then match the pattern against known causes like no-exit observer paths, run one discriminating check, and only retry after the check supports a changed plan.

How to stop an AI agent from burning tokens without progress?▼

Trim low-signal context such as duplicated plans and oversized logs, restate the real objective in one sentence, and shrink the task to one failing command or test. Escalate to a human if the failure is high-risk or externally blocked.

What are common AI agent failure patterns and their causes?▼

Common patterns include repeated identical commands from loop logic, context overflow from unbounded notes, ECONNREFUSED from wrong ports, 429 errors from retry storms, and stale file state from wrong cwd or branch drift.

When should I not use agent self-debugging workflows?▼

Do not use it for feature verification after code changes, where a verification-loop skill fits better, or for framework-specific debugging covered by a narrower skill. It also cannot enforce runtime behaviors the harness does not support.

Does agent introspection debugging require a specific project structure?▼

Yes, the frontmatter declares compatibility with a spec-kit project structure containing a .specify/ directory. The workflow itself is prompt-based and does not require external packages or scripts.