360 Secure SkillHub
360 安全技能中心

OpenClaw Self-Healing System

面向 OpenClaw Gateway 的四层自治恢复系统：L1 配置变更检测与重载，L2 看门狗检测 OAuth 过期、清理僵尸进程并指数退避重启，L3 连续失败后由 Claude Code 诊断修复，L4 通过 Discord/Telegram 升级给人工，含指标面板与安装脚本。

ramsbaby
ramsbaby
热度 73.41
最近更新 2026-05-11 11:30:56
分类：自动化依赖：代码执行 / 读写文件 / 第三方 API / 内部调用大模型应用场景：自动化脚本 / DevOps / 工作流

安装技能

复制下面的话术发给你的 Agent，它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill： npx secure-skills-cli add openclaw-self-healing
Metadata
name
openclaw-self-healing
version
3.1.1
description
4-tier autonomous self-healing and auto-recovery system for OpenClaw Gateway. Monitors gateway health, auto-restarts on crash, detects OAuth token expiry, kills zombie processes, and escalates to Claude Code AI for diagnosis when automated recovery fails. Use when your OpenClaw gateway crashes, stops responding, enters a restart loop, or needs automatic monitoring and recovery. Features watchdog, config validation, exponential backoff, Discord/Telegram alerts. macOS & Linux.
metadata
{"openclaw":{"requires":{"bins":["tmux","claude","jq"]},"install":[{"id":"tmux","kind":"brew","package":"tmux","bins":["tmux"],"label":"Install tmux (brew)"},{"id":"claude","kind":"node","package":"@anthropic-ai/claude-code","bins":["claude"],"label":"Install Claude Code CLI (npm)"},{"id":"jq","kind":"brew","package":"jq","bins":["jq"],"label":"Install jq (brew) - for metrics dashboard"}]}}

OpenClaw Self-Healing System

"The system that heals itself — or calls for help when it can't."

A 4-tier autonomous recovery system for OpenClaw Gateway, featuring AI-powered diagnosis via Claude Code. Tested in production on macOS + Linux.

Architecture

Level 1: config-watch        → Config file change detection + instant reload
Level 2: Watchdog v4.4       → OAuth detection, zombie kill, exponential backoff
Level 3: Claude Code Doctor  → AI-powered diagnosis & repair (30 min window) 🧠
Level 4: Discord/Telegram    → Human escalation with full context

What's New in v3.1.0

  • Complete healing chain fix — config-watch → Watchdog → Emergency Recovery now fully connected
  • Installer rewrite — single install.sh covers macOS (LaunchAgent) + Linux (systemd)
  • Watchdog v4.4 — OAuth token expiry detection, zombie process auto-kill, Exponential Backoff
  • Emergency Recovery v2 — persistent learning repo, reasoning logs, multi-model support (Claude Code + Aider)
  • Metrics dashboard — success rate, MTTR, trending analysis via tmux

Quick Setup

bash <(curl -fsSL https://raw.githubusercontent.com/Ramsbaby/openclaw-self-healing/main/install.sh)

Or install via ClawHub:

npx clawhub@latest install openclaw-self-healing

The 4 Tiers in Detail

Level Script Trigger Action
L1 config-watch.sh Config file change Validate + reload gateway
L2 gateway-watchdog.sh Process down / HTTP fail Kill zombie → restart → backoff
L3 emergency-recovery-v2.sh 30min continuous failure Claude Code PTY diagnosis
L4 emergency-recovery-monitor.sh L3 triggered Discord + Telegram alert

Configuration

All settings via environment variables in ~/.openclaw/.env:

Variable Default Description
DISCORD_WEBHOOK_URL (none) Discord webhook for L4 alerts
OPENCLAW_GATEWAY_URL http://localhost:18789/ Gateway health check URL
HEALTH_CHECK_MAX_RETRIES 3 Restart attempts before L3 escalation
EMERGENCY_RECOVERY_TIMEOUT 1800 Claude recovery timeout (30 min)

Verified Recovery Cases

  • OAuth token expiry — Watchdog v4.4 detects 401 in logs, restarts before agent dies
  • Zombie process — Preflight detects PID mismatch, SIGKILL + launchctl kickstart
  • Config schema error — openclaw doctor --fix auto-applied on exit_1 pattern
  • Level 3 triggered — Claude Code diagnosed and fixed broken config in < 15 min

Links

License

MIT — built by @ramsbaby + Jarvis 🦞