August 9, 2026
Bot drama, but make it secure
UnYOLO: Agent credential broker and policy engine for your GitHub account
This app says your AI helper can’t wreck your GitHub, but the comments are already side-eyeing it
TLDR: unYOLO is a tool that keeps AI assistants from getting full control of your GitHub account by forcing them through narrow, human-approved rules. Commenters liked the idea but instantly argued it may only solve the easy part, with one side warning about data snooping and another mocking its pipe-to-shell installer.
A new project called unYOLO is pitching itself as the grown-up in the room for AI coding assistants: instead of handing an agent the full keys to your GitHub account, it keeps the real login tucked away and only lets the bot do specific jobs you approve. In plain English, it’s a safety middleman. Want the agent to read code or push to its own little branch? Fine. Want it to rewrite your main branch and potentially torch your project history? Absolutely not—unless a human gives a temporary thumbs-up.
But the real fireworks were in the community reactions, where the mood swung from “finally, some brakes for AI” to “hold on, this only solves half the problem.” One of the sharpest comments cut right to the heart of the anxiety: sure, limiting what an agent can do is nice, but what about what it can see? That sparked the deeper fear underneath all the optimism: maybe the scariest part isn’t a rogue button press, it’s an overeager bot slurping up private data in the first place.
Then came the comedy roast. A commenter dunked on the project’s security vibe by pointing out the classic “download this install script and pipe it straight into your shell” move, which is basically catnip for skeptical developers. So yes, unYOLO wants to stop reckless AI behavior—but the crowd is already asking whether the tool is secure, complete, or just a very polished baby gate for robots
Key Points
- •unYOLO separates provider credentials from agents by using a broker process that executes approved operations on the agent's behalf.
- •The framework enforces authorization through a single JSON rules file loaded at startup, with deny rules taking precedence and invalid policy definitions preventing startup.
- •Requests follow a standard flow that authenticates the client, classifies the operation and target attributes, checks policy rules, and logs decisions without including secrets.
- •Requestable operations create durable approval records that operators can review through a protected inbox or Telegram, with approvals limited by time and use count.
- •The repository includes brokers for GitHub and Hugging Face, as well as sudo-broker for approved Unix commands, and each broker is isolated from other providers' credentials.