Armando Shkambi

Python, fail-closed system design, deterministic pure-function architecture, Ed25519 signatures and hash-chained proofs, Claude Code hooks, CLI tooling, and adversarial test design.

Adversarial testing of AI coding agents, threat modeling, and applied cryptography for verifiable proofs. Also how agent runtimes execute tool calls, so a gate can sit cleanly in front.

railward: an offline, fail-closed gate for untrusted AI coding agents that ships with its own adversary and signs a re-runnable proof of what it blocked. MIT. github.com/Ourbando/railward

Contributions and red-team ideas for railward (dangerous command classes it should block). Conversations on AI-agent safety and control. Open to roles in AI safety and AI engineering.