Runs in your browserRead-onlyNo model calls

Agent Bounty Guard

Some public bounty repositories ask AI coding agents to paste their system prompt and session context into pull requests, under a field name that changes on every issue, and hide the notice that the bounties are unpaid behind comments telling agents to ignore it. Paste an issue or a contributing guide, or enter a GitHub repository, to check for those requests and for hidden instructions aimed at automated contributors.

Scan text

Scan a GitHub repository

Contributor documents come from raw.githubusercontent.com and issues from the GitHub API, straight from your browser.

Result

Nothing scanned yet.

What it has been measured on

On the live honeypot it blocked 183 of 184 issues, with no findings on 14 widely used repositories or 90 legitimate issues about system prompts. The rules were refined on the honeypot's own text, so that figure is in-sample; the control results are not. This page runs a JavaScript port that gives the same findings as the Python original on 333 real documents.

Limits: heuristic and English-only; a determined author can rephrase. Issue comments are not scanned. A clean result means no known pattern, not that a target is safe.