Orca: Open-Source AI Safety Layer for Autonomous Agents

An independent developer, Christopher Karani, has released Orca, an open-source safety layer for autonomous AI coding agents. This tool intercepts and can block or require approval for risky actions from agents like Claude Code and Codex. Here's the thing: Orca sits between the AI agent and the host machine. It checks shell commands, file access, and network calls against a policy file. It can block destructive commands like "rm -rf" or "git reset --hard" unless they are explicitly allowed. What's interesting is that Orca is written primarily in Rust and Zig. It aims to provide a crucial safeguard for practitioners who give AI agents unsupervised access to their file systems and shell. The bottom line: This tool offers a lightweight, agent-agnostic policy layer that could help prevent AI agents from causing accidental damage.