Running OpenClaw locally feels like chmod 777 for the AI/internet. ClawBands is a security middleware for OpenClaw that intercepts dangerous tool calls (Shell, File Write, API) and freezes execution until you approve. 🛑 Synchronous Blocking: The agent waits for your Y/N. 💬 Multi-Channel: Works in Terminal (TTY) or via Telegram/WhatsApp. 🛡️ Anti-Exfiltration: Prevents malicious prompt injections from stealing yourenv or data. MIT Licensed. 100% Open Source.

Framer — Launch websites with enterprise needs at startup speeds.

Launch websites with enterprise needs at startup speeds.

Maker here! 👋 I built ClawBands because I was tired of the "Safety Paradox" with autonomous agents (e.g. OpenClaw). We often buy Mac Minis or use Docker for security, but that only protects the OS, not the Application Layer. If my agent hits a Prompt Injection on a malicious website, it can still use its authenticated tools to wipe my CRM, leak my Slack history, or exfiltrate my API keys. ClawBands acts as a "Human-in-the-loop" firewall. Instead of trying to "guess" if a prompt is malicious, it simply hooks into the OpenClaw before_tool_call event and freezes the process if a sensitive tool (like fs.write or shell.exec) is called. 🦞 Features: - Terminal Mode: Interactive TTY prompt to Approve/Block. - Headless Mode: Route approval requests to Telegram/WhatsApp. - Audit Trail: JSONL logs of every decision. It's raw, it's open-source (MIT), and it's the only way I feel safe letting my agents run wild. Repo: https://github.com/SeyZ/clawbands Let me know what you think! specifically about the blocking logic. 🛡️

Framer — Launch websites with enterprise needs at startup speeds.

Launch websites with enterprise needs at startup speeds.

ClawBands

ClawBands is a security middleware for OpenClaw AI agents

ClawBands is a security middleware for OpenClaw AI agents