🦞

Fix your OpenClaw
in one command

AI-powered diagnostic and repair. No SSH access needed. Runs locally, sends redacted logs, gets a fix script back.

v0.9.1 live Modernized diagnostics. Stricter repair safety. →
$ npx clawfix@0.9.1

Recommended — auditable source on npm and GitHub

Want to inspect before running? npx clawfix@0.9.1 --dry-run

Works on macOS, Linux, and WSL. Requires Node.js 18+.

✓ GitHub OIDC publish ✓ npm attestation verified ✓ 7-file allowlisted package
Release 0.9.1

Evidence before repair.

ClawFix now rejects incomplete diagnostics instead of turning uncertainty into repair advice. The release itself is signed, attested, and reproducible from public source.

Release notes →

Native contracts

Structured OpenClaw output is validated before issues are classified.

Ownership evidence

Port conflicts require proof of a competing listener—not merely an occupied port.

Fail-closed repairs

Repair validators reject partial output, timeout ambiguity, and unsafe catalog entries.

How It Works

1

Run One Command

The diagnostic script scans your OpenClaw installation. Config, logs, plugins, ports — everything checked in seconds.

2

AI Analyzes

Pattern matching catches known issues instantly. AI handles novel problems with deep analysis of your specific setup.

3

Review & Apply

You get a commented fix script. Read it, understand it, then run it. Nothing happens without your approval.

What It Detects

💀

Gateway Crashes

Port conflicts, process hangs, restart loops

🧠

Memory Issues

Mem0 silent failures, missing flush, broken search

🌐

Browser Automation

CDP port failures, extension loading, headless issues

🔌

Plugin Configs

Broken plugins, missing dependencies, wrong settings

⚡

Native Codex Harness

PI route drift, Codex home mismatches, fast tier gaps, timeout boundaries

💸

Token Waste

Excessive heartbeats, no pruning, bloated context

🍎

macOS Quirks

Metal GPU crashes, Apple Silicon issues, Peekaboo

🔧

Service Manager Crashes

launchd/systemd SIGTERM recovery, crash loops, backoff detection

👻

Zombie Processes

PID exists but port not listening — stale gateway detection

📜

Error Log Bloat

Chrome extension spam, handshake storms, 200MB+ log files

🐕

Gateway Watchdog

Recommends independent health checks to avoid launchd backoff gaps

🧵

Provider Prefix Typos

codex/gpt-5.4 vs openai-codex/gpt-5.4 — silent 403 + fallback loop on every cron

🎣

Discord Silent Drops

groupPolicy: allowlist with empty allowFrom — group messages disappear without logs

🔒

Plaintext Secrets in Config

Flags fields still inline that should be SecretRefs pointing at ~/.openclaw/.env

🪪

Invalid GH Token Override

Invalid GH_TOKEN env shadows a working gh login and breaks every GitHub-using cron

📡

Stale Paired Nodes

Endless skills-remote probe timeouts from a paired node with no host daemon behind it

🌊

Context Overflow

Session stuck >100 % of ctx window, auto-compaction failing — manifests as slow replies

🔐

FileVault Blocks Reboots

macOS — pre-boot prompt gates all services; any unattended reboot leaves the mac off-network

📦

Plist Stale Secrets

macOS — LaunchAgent EnvironmentVariables carries old secrets after a .env migration

Pricing

For now
Free

Every feature, for everyone

Pattern-matching scan, AI analysis, generated fix scripts — all free while we're figuring out what's worth charging for.

We may introduce paid tiers later (likely usage-based for heavy AI calls, or a hosted monitoring SKU). We'll announce before anything changes.

Security & Transparency

You're right to be skeptical of tools from the internet. Here's exactly what ClawFix does and doesn't do — verify it yourself.

🔍

Inspect Before Running

npx clawfix --dry-run shows exactly what data would be collected — sends nothing. Read the output. Decide for yourself.

🔓

100% Open Source

Every line on GitHub. CLI source, server code, diagnostic script — all public. Audit it.

🔒

Secrets Auto-Redacted

API keys, tokens, passwords — all replaced with ***REDACTED*** before anything leaves your machine. The env block is skipped entirely.

🚫

No File Contents Read

ClawFix checks if SOUL.md exists — it never reads what's inside. No chat history, no memory files, no personal data.

👀

Consent Required

The diagnostic asks [y/N] before sending anything. No data leaves your machine without you typing "y".

💾

Fix Scripts = Your Review

Fix scripts are saved to /tmp for you to read first. Every fix backs up your config. Nothing auto-executes.

📦 What Exactly Is Sent

✅ SENT (redacted)

  • • OS type, version, architecture
  • • Node.js and npm versions
  • • OpenClaw version
  • • Config structure (all secrets redacted)
  • • Recent error log lines
  • • Plugin names + enabled status
  • • Gateway status
  • • Hostname hash (8 chars of SHA-256)

❌ NEVER SENT

  • • API keys, tokens, passwords
  • • File contents (SOUL.md, memory, etc.)
  • • Chat history or messages
  • • IP address or real hostname
  • • Environment variables
  • • Personal data of any kind