Your AI agent runs third-party skills and MCP servers with full access to your files, keys and shell — no sandbox, no review. I scanned 96,096 published skills; 751 were malicious. PanGuard vets a skill before you install it, scans what you already have, and blocks hijack attempts at runtime. Free, MIT, fully on-device. Powered by 768 open ATR rules, already merged into Microsoft, Cisco, MISP and OWASP tooling. One command: npm install -g panguard && pga up
Ran the install in under a minute and it immediately flagged two skills I'd been using for months as sketchy. Wild that something this thorough is free and on-device.
Dial
751 out of 96k is a much higher malicious rate than I expected, that's a genuinely useful number to have public. for the runtime hijack blocking specifically, is that based on known attack signatures or are you also catching novel prompt-injection-style hijacks that don't match an existing rule yet?
Dial
solo builder, self-taught, four months in, and already merged into Microsoft/Cisco/MISP/OWASP tooling is a wild pace. one question on the runtime watcher piece specifically, does watching the agent at runtime add any noticeable latency to its actual work, or is it lightweight enough to just leave running in the background all the time
love that you put real numbers out there instead of hand-wavy claims, the 751 out of 96096 figure is the kind of evidence that makes this feel like a serious tool rather than a pitch. shipping it open source with the ATR rules already folded into MISP and OWASP is genuinely impressive execution.
@zzetcanla2en love you to man ,the 751 needs an asterisk. it was a snapshot, registry had moved by the next morning, so treat it as a floor not a census.
and the numbers split by corpus rather than being one figure. on the skill corpus the pattern layer does well, on prompt injection specifically (PINT, 850 samples) it only gets 63.6%. that gap is the reason there's a runtime guard and not just a scanner.
Macaly
751 malicous skills out of 96k is genuinely scary 😅 runtime guard for agents is so needed. congrats 👏
@petrkovacik thanks man, keep grinding
finally something that actually scans the skills i already installed instead of just trusting them. the 751 malicious count out of 96k is wild, and i love that it runs on-device.