Rlhf Feedback Loop

Rlhf Feedback Loop, also known as ThumbGate, is a local-first, open-source firewall and enforcement layer for AI coding agents. It intercepts and blocks dangerous or repeated tool calls (like 'rm -rf' or 'git push --force') before they are executed, using feedback-driven prevention rules derived from user thumbs-downs and prior session errors. Designed for developers and teams using agentic coding frameworks (like Claude Code, Cursor, Codex, and other MCP-compatible agents), it helps stop repeated mistakes, enforces compliance (including regulated-industry policies), and provides audit trails, saving money and improving reliability. No server is required; all enforcement happens locally on the user's machine.