En

Agent优化专家

Automated Diagnosis and Repair for Agent Execution Issues, Self-Healing Engine

Install & Use

Copy this prompt and send it to your AI assistant (Claude / Cursor / TRAE / Codex / WorkBuddy etc.) to auto-install:

Help me install this AI Skill: Agent优化专家.
It is used for: Automated Diagnosis and Repair for Agent Execution Issues, Self-Healing Engine
Full Skill content: https://321skill.com/skills/agent-diagnosis-repair-2/raw/index.md
Read that page and install it.

The prompt includes a link to the full Skill content. You can also view the full content.

This Skill addresses common pain points in Agent daily operations, such as frequent Cron job failures, tool invocation errors, unexpected workflow interruptions, and gradual performance degradation. In practice, developers often need to manually check logs, adjust configurations, and restart services. This Skill automates that process, providing a complete closed loop from detection to repair to learning.

Usage is straightforward: simply describe the issue naturally to your AI assistant (e.g., Claude, OpenClaw, or Hermes Agent), such as "Check why the Cron job failed" or "Perform a system health check." It will automatically execute a diagnostic decision tree, identify the root cause, and provide repair suggestions or directly execute safe fixes based on an action tier system (L0-L3). The entire process requires no manual script writing or memorization of complex commands.

It is ideal for teams or individuals who frequently need to maintain Agent stability, especially those who have deployed OpenClaw or Hermes Agent and are running multiple Cron jobs, tool invocation chains, or sub-agents. Operations engineers, backend developers, and agent developers can all benefit by reducing manual troubleshooting time and enhancing system reliability and self-evolution capabilities.

It is recommended to use this tool for scheduled daily checks or when failures occur. Note that it primarily focuses on Agent execution-level issues (e.g., timeouts, permissions, configuration errors) and is not suitable for business logic debugging, Lark content editing, or data query tasks. Additionally, its automatic triggering mechanism requires environmental support for heartbeats or scheduled tasks; otherwise, manual invocation is needed.

Key Features

Unlike traditional monitoring and alerting tools (e.g., Prometheus + Alertmanager), it not only detects anomalies but also automatically performs diagnosis, repair, and validation, and possesses self-evolution capabilities—each repair contributes error patterns to the `learnings/` directory, with periodic updates to `fix-templates`, forming a continuous improvement closed loop.

Limitations

Requires an OpenClaw or Hermes Agent environment and a heartbeat mechanism for automatic triggering. It does not support troubleshooting at the business logic level, being limited to diagnosing and repairing the Agent execution environment (Cron, tools, sub-agents, system resources).

FAQ

Can this Skill automatically fix all Cron failure issues?

It handles most common causes, such as schedule expression errors, timeouts, disabled states, etc. For complex business logic errors, it will still log the issue and recommend manual intervention.

How does it ensure the safety of repair operations?

It employs an action tier system (L0-L3): L0/L1 actions are executed automatically (e.g., reading status, cleaning temporary files), L2 actions prompt for user confirmation (e.g., modifying cron, restarting services), and L3 actions require explicit user authorization (e.g., modifying core configurations, deleting data). Automatic modification of core files like `SOUL.md` is prohibited.

Installation guide for AI assistants

If your AI coding assistant (Claude Code, Cursor, TRAE etc.) can see this page, send it this message to auto-install:

Visit https://321skill.com/skills/agent-diagnosis-repair-2/raw/index.md to read the original Skill definition (Markdown format) for Agent优化专家, and install it according to the instructions.