ProjectIntent Over Keywords: From Stitched Embeddings to a Production Fashion Agent · Module 21: Security: threats, guards and limits · part 23 of 26
Part 23 · Module 21: Security: threats, guards and limits
Module 21: Security: threats, guards and limits
24 min read·
About 50 minutes · Needs: Module 12 (the high-level design: nginx in front of the API), Module 13 (the chat API contract, sessions and settings), Module 17 (grounded generation) · Code: part3-styleme/backend/src/styleme/uploads.py, limits.py, ratelimit.py, api.py, agent/guards.py, part1-video-rag-stitched/src/extractors/media.py, src/ui/app.py, part2-video-rag-unified/src/api/common.py, part3-styleme/frontend/nginx.conf.template, frontend/src/security.test.tsx
Where this fits
The parts this module guards (blue, the guards themselves green): nginx and the StyleMe API at the front door, the planner and generator prompts, the web app that renders replies, and the video labs' ffmpeg steps and UIs.
Earlier modules built features. This one asks how each feature can be abused, and shows the code that stops it. The highlighted parts are where attacks land: nginx and the StyleMe API at the front door, the planner and generator prompts that read text written by strangers, the web app that displays model output, and the ffmpeg steps and Streamlit UIs of the two video labs.
The rest of this project is yours to keep
This project is bought on its own, once, and stays readable afterwards, including the parts added to it later.