soil-teaching-deck
Pass
Audited by Gen Agent Trust Hub on Oct 27, 2026
Risk Level: SAFEDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [DYNAMIC_EXECUTION]: The skill dynamically generates Python and Node.js scripts (such as 'geo_custom.py' and 'geo_to_b64.py') at runtime to facilitate geometry rendering and file conversion. These scripts are executed locally to produce intermediate assets and the final presentation file.
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external educational materials and existing presentation files. This ingestion of untrusted data creates a potential surface for indirect prompt injection, although the multi-stage structured workflow provides logical boundaries that help mitigate accidental instruction obedience.
- [COMMAND_EXECUTION]: The skill invokes various shell commands to install necessary dependencies, search for rendering components in local skill directories using 'find', and run system utilities like 'soffice' and 'pdftoppm' for PDF and image conversion.
- [EXTERNAL_DOWNLOADS]: The instructions involve downloading and installing standard packages from official registries, such as 'pptxgenjs' via NPM and 'markitdown' via PyPI. These are legitimate dependencies required for the skill's primary functions.
- [DATA_EXFILTRATION]: Analysis confirms that the skill only performs local file operations, reading from and writing to the current working directory. No unauthorized network transmissions or attempts to access sensitive system files (like credentials or SSH keys) were detected.
Audit Metadata