Instead of just filtering malicious prompts, I implemented role-based policies with runtime input validation that can scale across combinations of different AI tools (GitHub, Stripe, Linear, etc.).
All the code referenced in the blog is in the examples/supabase_demo folder.
I welcome your feedback — especially from folks working with AI toolchains or security.
anand-tan•11h ago
https://news.ycombinator.com/item?id=44502318