About
I read code, find where a trust boundary breaks, and prove the bug against the real running software before I disclose it.
Most of what I find sits in developer tooling. MCP servers, editor plugins, self-hosted apps, browser extensions. The pattern that repeats is untrusted synced content, or an open local port, reaching something it was never meant to.
Everything gets a working proof-of-concept and a coordinated report. A clean audit with no finding is a legitimate result too, and I report it that way.
2CVEs assigned
13advisories published
7certifications
CVEs
WolfStack9.8 Critical
A cluster-authentication secret is hard-coded into every build and published, so any unauthenticated client reaches full remote code execution.
emlog6.8 Medium
The AI-assistant exec endpoint has no CSRF protection, so a visited page can drive SQL execution and take over an admin account.
Published Advisories
A CVE has been requested for each and is pending assignment. Additional findings are in coordinated disclosure.
Certifications
Contact