The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The FeaturePipeline class in references/feature-engineering.md uses pickle.load() to restore pipeline objects from files. This is a known security vulnerability as pickle can be exploited to execute arbitrary code during deserialization.\n- [REMOTE_CODE_EXECUTION]: In references/training-pipelines.md, the Trainer.load_checkpoint method calls torch.load() to restore model state. By default, torch.load uses the unsafe pickle module, which presents a risk of arbitrary code execution if a user loads a malicious checkpoint file.\n- [COMMAND_EXECUTION]: The skill provides a shell script template (launch_distributed.sh) in references/training-pipelines.md that executes torchrun. This pattern facilitates direct shell command execution for orchestrating distributed training processes.

ml-pipeline