model-pruning

Warn

Audited by Gen Agent Trust Hub on Jun 29, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill downloads external research code from third-party GitHub repositories: https://github.com/locuslab/wanda and https://github.com/IST-DASLab/sparsegpt.
  • [COMMAND_EXECUTION]: The installation instructions require executing pip install -r requirements.txt and pip install -e . within the downloaded directories. This results in the execution of installation scripts provided by these external sources.
  • [PROMPT_INJECTION]: The skill presents a potential surface for indirect prompt injection during model calibration.
  • Ingestion points: calib_data parameter in wanda_prune (found in SKILL.md and references/wanda.md).
  • Boundary markers: Not present; the text is tokenized and fed directly to the model.
  • Capability inventory: The skill can write to the file system using model.save_pretrained.
  • Sanitization: None; calibration samples are processed without filtering.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jun 29, 2026, 12:55 AM
Security Audit — agent-trust-hub — model-pruning