model-pruning

Pass

Audited by Gen Agent Trust Hub on Sep 17, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill fetches source code for model pruning implementations from the Locus Lab and IST-DASLab repositories on GitHub.
  • [COMMAND_EXECUTION]: The skill executes standard shell commands for environment setup, including git clone for code retrieval and pip install for managing dependencies such as torch, transformers, and accelerate.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests external text data from sources like the C4 and WikiText datasets to calculate activation statistics during the pruning process.
  • Ingestion points: External text samples are passed into the wanda_prune and SparseGPT.prune functions via the calib_data parameter.
  • Boundary markers: No explicit delimiters are used for the calibration data, as the text is used for statistical analysis rather than instruction following.
  • Capability inventory: The skill possesses the ability to write to the file system to save compressed models and perform network operations during setup.
  • Sanitization: The input text is processed directly to compute activation norms without secondary sanitization filters.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 17, 2026, 07:53 PM
Security Audit — agent-trust-hub — model-pruning