albert-weasker/niubi_guard

Niubi Guard is a free, open-source tool that helps GitHub project maintainers defend their repositories against spam, harassment, and coordinated abuse campaigns. It was built in response to a pattern maintainers have reported: waves of hostile issues, repeated copy-paste accusations, and reputation-pressure attacks that normal moderation tools are not designed to handle. The tool works by scanning your repository's issues and comments and flagging content that matches abuse patterns. Detection can rely on keyword rules you define, a blocklist of specific usernames, allowlists that protect trusted contributors, or an AI model of your choosing. For AI detection, Niubi Guard connects to any OpenAI-compatible API, meaning you supply the API key, model name, and custom prompt. The model returns a structured verdict (malicious or not, a confidence score, and a reason), and you set the confidence threshold at which the tool acts. All destructive responses, such as deleting comments, closing issues, locking threads, or blocking users, are turned off by default. You must explicitly enable each action in a configuration file and then pass an apply flag when running the tool. Without that flag, it only prints what it would do, making dry-run the safe default for new users. Niubi Guard ships with three ways to run it: a command-line tool installable from npm, a web-based dashboard built with Next.js that lets you configure everything through a browser without touching files, and a Docker image for self-hosted setups. A free hosted version is also available at niubistar.com/guard for maintainers who do not want to self-host. The interface supports English and Simplified Chinese. The project is licensed under Apache-2.0 and welcomes contributions, including new attack samples, false-positive examples, prompt improvements, and eventual GitHub App or GitHub Actions support. The roadmap shows plans for a review queue, community threat feeds, and a full GitHub App integration in a later version.