Show HN: Fence – Jiminy Cricket for AI coding agents
TL;DR
hoop.dev founder Andrios R. posted Fence on Hacker News, an open-source tool for AI coding agents that came out of an internal 20-percent side-project sprint. Fence is meant to stop catastrophic shell commands before Claude Code or Codex can execute them, including rm -rf-style variants aimed at home directories. The pitch: Fence is not a simple denylist. It claims to infer command intent. The post does not yet provide evidence on accuracy, false positives, bypass resistance, or real-world incidents.
Nauti's Take
For teams using AI coding agents, this is worth testing as an extra checkpoint, but it is not a security model yet. Run it against your own risky command examples, wrapper scripts, and legitimate deploy flows first.
The useful question is whether it catches real mistakes without turning normal engineering work into a stream of false alarms.
Briefingshow
AI coding agents increasingly get terminal access, write files, and run commands. A bad prompt, model mistake, or misunderstood task can damage a local machine before a human notices. An intent filter before execution can be a useful second brake, but it does not replace sandboxing, least privilege, and backups.
Sources
