// the find
SkalskiP/awesome-chatgpt-code-interpreter-experiments
Awesome things you can do with ChatGPT + Code Interpreter combo 🔥
A documentation repo (no code, just a README full of screenshots) cataloging jailbreaks and tricks for OpenAI's ChatGPT Code Interpreter circa mid-2023 — installing pip packages offline via .whl files, running Deno/JS instead of Python, pulling the system prompt, and some CV experiments like Haar Cascade face tracking done without internet access. Useful for anyone curious about how far you could push that specific sandbox before OpenAI patched around it.
The .whl-based pip install workaround and the 'ask nicely, more than once' social-engineering pattern are concrete and reproducible, not vague claims. The CV experiments (color-based object tracking, Haar Cascade face detection, IoU-based filtering) are legitimate from-scratch techniques for working around a sandbox with no pretrained models, and show real problem-solving rather than just prompt copy-paste. The system prompt extraction is a genuinely useful historical artifact for anyone studying how OpenAI scoped that tool.
Last commit is from December 2023 — OpenAI has since rebranded Code Interpreter as 'Advanced Data Analysis' / enabled internet access in some contexts, so several of the 'limitations' and workarounds here are stale or no longer apply, and the repo gives no indication of what still works. There's zero code in the repo itself; everything is screenshots embedded in a README, so you can't actually run or verify any of it, you're trusting images. It's also scoped entirely to one proprietary product's sandbox quirks rather than a general technique, so its shelf life is inherently short and it has had no activity in nearly two years to show whether it's been kept current.