raktim-yoddha/pr-guardian ? reverse-engineered prompt

Build me a web app called PR Guardian that watches GitHub pull requests for a connected repository and acts like an automated first reviewer.

I want users to be able to sign up, log in, connect their GitHub account, pick a repo, and create an agent for it. Once that’s set up, the app should ingest the repo and its issues, then automatically review incoming PRs, check for obvious prompt injection, spam, and malicious code, and only then approve clean ones with a rewritten reviewer friendly summary. If a PR looks bad, it should decline it and flag the account so repeated abuse can be tracked.

Please include a dashboard where I can see live PR progress, the decision for each PR, layer by layer results, and a list of flagged accounts with a way to unban them. Support dark and light mode, GitHub webhook based updates, and a manual sync option for missed PRs. Use a CPU friendly approach for retrieval and let users bring their own LLM key, with encrypted storage for those keys. If you need current docs for any libraries, look them up online.