skillxray exists to read hostile input: agent skills that may contain prompt injection, obfuscated commands, or worse. It is static by design - it reads files, matches patterns, and prints a report. It never executes anything from the skill it's scanning, never imports it, and never talks to the network.

That makes the scanner itself the attack surface. A skill crafted to crash the parser, to hide a finding from it, or to smuggle terminal escape sequences into the report so they run in your terminal when you read it - any of those is a vulnerability in skillxray, and exactly the kind of report we want. A missed detection (a payload the scanner should reasonably catch but doesn't) is welcome as a regular issue with a sample.

Please don't open a public issue for security problems. Use GitHub's private reporting instead:

https://github.com/munzzyy/skillxray/security/advisories/new

Include what you found, how to reproduce it, and the impact you'd expect.

Fixes land on the latest tagged version; there's no backport policy.