Teams shipping LLM products face a new class of bugs: prompt injection, jailbreaks, data leakage. The testing tooling for them is mostly ad-hoc or locked inside vendor red-team teams. Smaller teams can't run that kind of program on their own.

Adversec exposes a small, focused API. You point it at your LLM endpoint, it runs a battery of adversarial probes, and you get back a structured report of what broke and why. The probe library is the core — categorized by attack type, versioned, and designed to evolve as new jailbreaks surface in the wild.

By the time I shipped it, Claude had already picked up enough built-in defenses and self-evaluation capability that a naive user might ask, "can't I just ask Claude to red-team itself?" Sometimes, yes. But having a purpose-built tool with a stable API, versioned probe sets, and side-by-side comparisons across models is a different thing than one-off prompting. It was also my first full-stack API — auth, rate limits, structured errors, docs — and those skills transferred to every project since.

I designed a crewneck for a theoretical Adversec merch drop — dark, minimal, the kind of thing a security engineer might actually wear. But the honest truth is Adversec doesn't have the audience or user base to justify a run. Maybe one day, if the platform finds its footing. Until then, it's a design sitting in the public folder as a reminder that shipping the product comes first — the merch is just a nice idea for later.