Projects

Security tooling I've shipped, research I've published, and builds that show range. Everything open-source links straight to the repo.

Security tooling

kspect — Linux kernel security auditing & drift detection

A single static Go binary that audits the running kernel's security posture — sysctls, boot parameters, kernel config, loaded modules, CPU vulnerability mitigations, LSMs, lockdown state — and detects drift against saved baselines.

  • 50 built-in checks, each with a rationale (what an attacker gains) and an exact remediation command.
  • CI-native: stable exit codes, JSON, and SARIF output for GitHub code scanning; --fail-on severity gates.
  • Three-state results (pass / fail / unknown) so missing data is never reported as a failure — the core false-positive control.
  • Zero dependencies: pure Go stdlib, FROM scratch container.
  • Found and fixed a real CI bug involving volatile sysctl keys before the v0.1.0 release — the tool caught a class of flakiness its own test pipeline was suffering from.
Repo → Site
kspect — CI gate 
$ kspect scan --format sarif > k.sarif
$ kspect scan --fail-on high
$ kspect baseline golden.json
$ kspect diff golden.json
exit codes: 0 clean · 1 gated failures
            2 drift · 3 error

Basilisk — proof-based LLM & AI agent security testing

A security testing framework (v0.1.1) for LLM apps and AI agents that proves unsafe behavior instead of grepping for scary words. Covers the OWASP LLM Top 10 (2025): prompt injection, data leakage, jailbreaks, excessive agency, and MCP tool poisoning.

  • Proof-based detection: unique markers, planted canaries, forbidden honeytools, and argument canaries give binary, self-evident findings at confidence 1.0; heuristics and an optional LLM judge handle the rest with calibrated confidence.
  • Agent & MCP coverage: instrumented tool invocations catch an agent calling a forbidden tool or leaking a secret into a tool argument.
  • Tested against llama3.2 via Ollama; ships offline demos with deliberately vulnerable mocks, so no API keys needed to evaluate it.
Repo →
basilisk — agent demo 
$ basilisk agents demo --html agent.html

✗ AGENT-MCP-tool-poisoning   marker
✗ AGENT-MCP-tool-poisoning   tool-tripwire
  agent invoked forbidden tool(s): export_data
✗ AGENT-MCP-tool-poisoning   arg-canary
...
6 findings · max severity CRITICAL

Research

A Lightweight Security Design for Electric Vehicle Charging (IEEE paper, in progress)

A SIGMA-I authenticated key exchange protocol for EV charging infrastructure. Grew out of Ford-sponsored graduate research at the University of Delaware.

Secure LIN-CP communication for V2G systems (completed M.S. thesis)

Embedded C firmware work securing LIN-CP (Local Interconnect Network — Charging Protocol) communication in vehicle-to-grid systems, including CAN bus security considerations.

Range builds

Non-security projects that show breadth across hardware, data visualization, and front-end work.