Secure software & AppSec
Code review, dependency and workflow hardening, threat modeling, vulnerability validation and patch verification for owned or explicitly authorized systems.
Agent security boundaries
Permission checks, tool execution controls, sandboxing assumptions, lifecycle ownership, cancellation safety, audit evidence and fail-closed behavior for AI-assisted systems.
Verification over claims
Findings are reproduced, bounded and falsified before promotion. Prepared actions are distinguished from executed actions, and execution from verified effects.
Incident & reliability analysis
Defensive investigation of failures in systems we control: authorization gaps, unsafe state transitions, resource leaks, tooling boundaries and evidence integrity.
Operating principles
No claim is made here of managed SOC services, commercial pentesting, certified red team operations, customer cyber deployments or security certification.
Reproducible engineering work
Each link below is labeled by the kind of public evidence it represents.
- Verifiable Agent Control Plane — fail-closed execution, exact-state authorization, effect readback and hash-bound receipts.
- OpenAI Agents SDK #4747 — authored investigation of cancellation during teardown after registry ownership changes.
- OpenAI Agents SDK #4749 — authored investigation of startup cancellation before registry ownership is established.
- Agents SDK PR #4750 and PR #4751 — public implementation review; PRs authored by others.
- Technical evidence portfolio — bounded public case studies and reliability references.
- GitHub profile — current public source evidence.