ClawSecure — 3-layer security audit with 55+ OpenClaw-specific threat patterns

Container sandboxes break agent utility because agents need filesystem, network, and shell access to do real work. But running agents unsandboxed on personal machines exposes users to file deletion, credential theft, and data exfiltration. This tool sits between the agent and the OS, performing behavioral analysis on every action (file writes, network calls, shell commands) and blocking destructive patterns through configurable policies, without restricting where the agent runs.