File and model safety scan
Tell me a file and I will say whether opening it is dangerous: pickle/torch opcode walk, Keras Lambda bytecode, zip-slip, packed blobs.
Acerca de este servicio
Give me any file from an untrusted source - a checkpoint (.pt/.pkl/.joblib), a .keras/.h5 model, an archive, or an unknown binary. You get a written verdict (SAFE / UNKNOWN / SUSPICIOUS / DANGEROUS) with every finding and the exact reason, plus a reproducible scan script. .pkl and .pt are opcode streams: torch.load() on an untrusted checkpoint executes code while reading the file, before any check you could write. A Keras Lambda layer is marshalled Python bytecode in config.json, unmarshalled on load. I read the stream rather than grepping for strings.