Armur-Ai/byakugan ? reverse-engineered prompt
Reverse engineered prompt
Build me a command line tool that can statically scan machine learning model files for hidden code execution risks before anyone loads them.
I want it to understand common model formats like PyTorch pickle based files, Keras files, plain pickle, ONNX, GGUF, and safetensors, then give a simple verdict like safe, warning, or dangerous. It should be able to scan one file or a whole folder, print a clear human readable report, and also support JSON and SARIF output for CI. If something looks dangerous, it should explain why in plain English, like suspicious pickle opcodes, unsafe imports, embedded code, or format policy violations, and return a nonzero exit code so it can fail a pipeline.
Also add a way to inspect the file more deeply, showing the risky points inside the serialized data, and a basic policy file so teams can require safe formats only. If you need to look up current format details online while building it, go ahead.
Are you gonna build this?
make sure you review the code using coderabbit