soia-pkm-transform-article-ppt

Pass

Audited by Gen Agent Trust Hub on Jul 22, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill utilizes a local Python script, scripts/media_bundle.py, to manage the creation and validation of presentation media bundles. Technical review of the script confirms it relies solely on standard Python libraries to perform file parsing and validation tasks.
  • [EXTERNAL_DOWNLOADS]: Documentation references the installation of notebooklm-py and the addition of vendor-specific skills through npx skills add. These resources are either associated with the vendor soia-team or well-known services like NotebookLM, representing standard integration procedures.
  • [DATA_EXFILTRATION]: The skill contains a robust privacy mechanism in scripts/media_bundle.py. It uses a specific regular expression (BANNED_TEXT) to scan generated PPTX files and block the inclusion of sensitive local information, such as system user paths (e.g., /Users/) or internal runtime metadata.
  • [PROMPT_INJECTION]: Because the skill processes external content like URLs and Markdown articles, it has a surface for indirect prompt injection. The skill mitigates this through 'Quality Gates' and explicit instructions to verify that all output is strictly grounded in the provided source material.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 22, 2026, 10:12 AM
Security Audit — agent-trust-hub — soia-pkm-transform-article-ppt