Team Idea
An open-source agent that audits any repo shipping AI features, scores it 0–100 across Design / Ship / Run / Secure / Test (literally the congress program turned into a metric), finds what's dangerous — injection-prone prompts, unguarded tool permissions, missing evals, secrets leaking into LLM calls — and opens real fix PRs: guardrail middleware, eval harnesses, CI gates. Built on your proven stack (Next.js 16 + TS agent loop + GLM-class tool calling), and the stage demo is bulletproof: paste a real repo, live score, live PR, audience QR submissions, cached fallback repos if the wifi dies.
Team Members
Shyam Desigan
impactquadrant655