CodeArmor MVP
An AI security scanner that lives inside VS Code and checks for 40+ classes of vulnerability.
The problem
Security review usually happens long after the code is written, in an audit nobody enjoys. The scanners that do run in the editor tend to fail in one of two ways: they miss the context around a line, or they bury you in warnings without saying how bad things are overall.
What we built
Four of us built CodeArmor as a ten-day MVP. Eleven specialized agents scan in parallel across more than 40 vulnerability categories, ArmorClaw checks each finding before it ever reaches a developer, and a scoring pipeline turns what survives into one number you can watch move. Results appear in a VS Code extension, and a web dashboard gives an admin the same picture across a team. I owned the backend and the AI integration.
- Eleven scanning agents running in parallel over 40+ vulnerability categories
- ArmorClaw validation layer that filters findings before a developer sees them
- Scoring pipeline feeding both the editor and the dashboard
- VS Code extension plus a web admin panel with role management
- Backend APIs, JWT auth, PostgreSQL with Drizzle ORM, and scan orchestration
My contribution
What I owned on the team
Backend developer, AI integration, and security architecture. I designed the APIs, wired up the scanning workflow, implemented JWT auth, modeled the PostgreSQL schema with Drizzle, built scan orchestration and the scoring logic, and worked with the extension developer on the VS Code side.
How it turned out
We had a working MVP at the Grand Finale: start a scan from the editor, get back a filtered, scored set of findings. Ten days from empty repository to that.