Pseudonymous solo research lab for AI/LLM security, detection engineering & OSINT tooling
Working artifacts, not promises. The five featured project repositories are MIT-licensed.
Python 3.12 · Sigma · MCP · OSINT · MITRE ATT&CK · Claude Code
Solo researcher working on AI/LLM security, detection engineering, and OSINT tooling. I publish small, security-focused open-source artifacts I actually use: vulnerable-by-design labs, a client-side secret scanner, Sigma detection content, OSINT trust envelopes, and a deny-by-default desktop-automation MCP server. Several core tools are standard-library-only. Bug reports and detection rules occasionally land upstream.
Vulnerable-by-design MCP server for learning object-level / cross-tenant authorization (BOLA/IDOR) bugs, paired with a hunt checklist and CI that keeps the vulnerable-fixture set honest.
Offensive & defensive AI/LLM security tools, labs, and CTF write-ups. Zero-dependency Python.
100% client-side secret scanner. Paste code or drop files; nothing leaves your browser (zero upload).
294 Sigma detection rules across 16 ATT&CK tactic categories, none marked `stable` — sigma's `status:` field used literally rather than aspirationally. Ships 3 MCP tools + 3 Claude Code skills.
Per-source epistemic ceilings for OSINT results: honest verified / inferred / heuristic / unverified envelopes. Zero-dependency Python.
desktop-automation-mcp Apache-2.0
Deny-by-default, per-action-gated Windows window MCP server. Every capability (observe, click, type, close, drag) is a separately granted permission, and every effect is re-verified against the live target (identity, focus, occlusion) right before it runs. It hands the client pixels and window-relative coordinates, never on-screen text, so screen content cannot be mistaken for instructions. UIA-less canvas apps are supported through versioned, hash-verified coordinate profiles.
Credited reporter on published GitHub Security Advisories:
- Pelican: application API mounts ACL-scope bypass, CWE-862/863 (GHSA-43h8-3896-wqv5)
- Pelican: Filament suspend-all / unsuspend-all missing authorization (GHSA-4wxv-r46p-w2f9)
- Jexactyl: free-billing flow missing ownership check, CWE-862 (GHSA-9xwv-p7r5-5h5p)
- OneUptime: incoming-call-number resend-verification-code missing ownership check — the sibling channel the CVE-2026-30959 fix sweep left uncovered, CWE-862 (GHSA-wc96-jm46-37hh)
Vendor security acknowledgement:
- zn: high-severity control-plane authorization gap on policy and consensus handlers, where non-admin tenant keys could modify policy or consensus state through missing function-level checks (SEC-2026-01, credited on the usezn.com Security Hall of Fame)
Detection templates merged into projectdiscovery/nuclei-templates:
- CVE-2026-26190 Milvus, unauthenticated metrics port (#16333)
- CVE-2026-25527 changedetection.io ≤ 0.52.9 (#16346)
- CVE-2026-33476 SiYuan ≤ v3.6.1 (#16335)
- CVE-2026-31831 Tautulli ≤ 2.16.1, unauthenticated path traversal (#16345)
0 of 294 sigma rules marked stable · no open CodeQL alerts across 5 featured repos (checked 2026-09-20) · MIT-licensed featured projects · standard-library-first where it makes sense