Hey yall, I’ve been trying to build autonomous AI agents using MCP lately, but I noticed a destructive issue: it’s incredibly easy to accidentally give an agent full access to your root directory, leak hardcoded API keys in configs, or forget manual approval gates on destructive tools. Since I couldn't find a quick way to solve this, I spent my past week building a lightweight Python CLI and GitHub Action called agentlint . It statically scans your configs for: Broad filesystem expoes...