model-pruning

This SKILL.md and accompanying code examples implement legitimate model pruning techniques (Wanda, SparseGPT, N:M) and are consistent with their stated purpose. There are no signs of malicious behavior or covert exfiltration. Main concerns are operational: the code modifies model weights in-place (risk of accidental data loss if users do not backup originals), several example helper functions/identifiers are undefined or missing imports (could lead to runtime errors), and installing external GitHub code requires standard supply-chain trust. Recommend users: (1) back up original checkpoints before pruning, (2) review and test example helper functions, (3) verify and pin upstream GitHub commits before pip installing in production environments.