agent-failure-introspection

Diagnose failed agent workflows and recommend fixes using Python scripts.

3|Updated Jun 8, 2026

Search Tags:#workflow #diagnosis #project analysis #agent failure #fix recommendation

Ready to run

npx skills add https://github.com/pax-k/pax-fullstack-harness --skill agent-failure-introspection

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: agent-failure-introspection
Source: https://github.com/pax-k/pax-fullstack-harness/tree/main/skills/agent-failure-introspection
Command: npx skills add https://github.com/pax-k/pax-fullstack-harness --skill agent-failure-introspection

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires scripts/diagnose_agent_failure.py, and includes scripts (resource) components.

What problem does it solve?

Diagnoses and resolves failed or unreliable agent workflows, identifying root causes and suggesting fixes.

Core Features & Use Cases

Failure Classification: Identifies bad assumptions, missing observations, tool misuse, and other failure patterns.
Evidence Collection: Captures failing command output, browser logs, and task plan details.
Fix Recommendation: Suggests durable fixes for harness or project workflow issues.
Use Case: After a harness run failure, use this skill to diagnose the issue and recommend a fix.

Quick Start

Run the agent-failure-introspection skill on the failed harness run to diagnose the issue.

Frequently Asked Questions about agent-failure-introspection

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

FAQPage Schema

How do I diagnose a failed agent workflow?▼

Diagnosing a failed agent workflow involves running a Python analysis script that classifies failure patterns like bad assumptions and tool misuse, then collecting evidence such as failing command output and browser logs.

What causes agent workflows to fail during a harness run?▼

Agent workflows fail during a harness run due to bad assumptions, missing observations, tool misuse, and harness or project workflow issues, which are identified through systematic evidence collection and failure classification.

How do I get fix recommendations for project workflow issues?▼

Fix recommendations for project workflow issues are generated by executing a diagnostic script that analyzes the failed harness run and suggests durable fixes for harness or project code problems.

Do I need Python to run agent failure introspection?▼

Yes, you need Python installed because the diagnosis and project code fix recommendation process relies on executing Python scripts to analyze the failed harness run and identify root causes.

What is the best way to capture evidence from failed agent workflows?▼

The best way to capture evidence from failed agent workflows is to use a diagnostic skill that automatically collects failing command output, browser logs, and task plan details for post-failure analysis.