mcp-firewall: A better policy-engine for CLI agents

I created a more granular policy-engine for cli agents (at the time of writing only claude code and github copilot cli are supported) Each agent implements its own halfway solution to allow/deny specific commands or tools (e.g. globs, prefix-only, etc.). Funnily enough the code is not vibe coded, almost at all (hence the use of the ai tag instead of vibecoding). I used an agent to generate the initial POC but there's almost 0 agent lines left. Comments