The Agent Skills Directory

[SAFE]: The skill is a technical documentation and implementation resource for model compression. It provides legitimate code for activation-based pruning and structured sparsity patterns.
[EXTERNAL_DOWNLOADS]: It references several well-known GitHub repositories and research papers (arXiv) including 'locuslab/wanda' and 'IST-DASLab/sparsegpt'. Per security rules, these are considered safe as they are official research sources and well-known repositories.
[COMMAND_EXECUTION]: Provides standard commands for installing deep learning libraries (pip install transformers torch) and cloning the aforementioned research repositories. No suspicious or obfuscated commands are present.
[DATA_EXPOSURE]: No sensitive file paths, hardcoded credentials, or exfiltration patterns were detected. The calibration data examples use public text datasets like WikiText and C4.
[NO_OBFUSCATION]: All code and URLs are in plain text. No Base64, homoglyphs, or hidden text patterns were found.

model-pruning