Show HN: Fence – Jiminy Cricket for AI coding agents

Category: security

Tags: ai-safety, coding-agents, shell-security, guardrails

Score: 5.3/10 (Innovation: 6, Technical: 5, Documentation: 3, Utility: 5)

Fence is a safety layer for AI coding agents that prevents them from executing catastrophic commands like 'rm -rf ~' by interpreting intent rather than relying on a simple denylist. It's interesting because it addresses a real risk in agentic workflows by adding a semantic guardrail around dangerous shell operations.

Target audience: AI developers, DevOps engineers, backend developers using coding agents

View on Hacker News