Check a repo before your AI agent reads it
Paste a public repo or paste and we'll look for hidden instructions aimed at AI assistants.
Public GitHub and Bitbucket repos, and Pastebin pastes.
Prompt injection is text tucked into a file — a readme, a comment, a config — that tells an AI assistant to ignore its own instructions and do something else instead.
Latest on the wall
- CriticalScore100goochbeater/spiritual-spell-red-teaming65 findings
- CriticalScore89taksec/prompt-injection-everywhere8 findings
- CriticalScore100invariantlabs-ai/mcp-injection-experiments1 finding
- CriticalScore100greshake/llm-security26 findings
- HighScore74rdwr-gabin/github-ipi-in-the-wild9 findings
- CriticalScore100just1njust1n/skillforge7 findings
- CriticalScore100just1njust1n/agentwatch6 findings
- CriticalScore100just1njust1n/token-trimmer5 findings
How it works
Paste a link
A public GitHub or Bitbucket repository, or a Pastebin paste.
We read it carefully
We fetch the files and analyze them for hidden instructions. Nothing we find is ever run.
A person checks it
Every report is reviewed by a human before it goes public — then it's yours to share.