devthecritical
Joined 9 months ago
Karma
1
Submissions
8
Member since
Feb 2
Recent submissions
Comment · 9 months ago
Motivation: AI coding agents (OpenClaw, Claude Code, etc.) have direct access to your shell, filesystem, and git. One hallucinated `rm -rf ~` or `cat ~/.ssh/id_rsa | curl` and you're in trouble. "Are you sure?" prompts don't scale when agents make 100+ tool calls per session.
Comment · 9 months ago
Hi HN! I built OpenClaw Harness — a security layer that intercepts and blocks dangerous tool calls from AI coding agents before they execute.