uv-create-paper-exercises
Pass
Audited by Gen Agent Trust Hub on Jun 27, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes local Python scripts to scaffold the exercise environment (
scripts/scaffold_exercise_pack.py) and to run exercise checkers and tests within generated sub-skills. - [EXTERNAL_DOWNLOADS]: Instructions guide the agent to download research papers (PDFs) and clone external Git repositories into a local directory (
$HUMAN_MATERIAL_PATH/.references/) for use as reference material. - [PROMPT_INJECTION]: The skill processes untrusted external data (research papers) and uses that information to generate exercise instructions, code solutions, and sub-skills, creating a surface for indirect prompt injection.
- Ingestion points: Research paper content (arXiv URL or PDF files).
- Boundary markers: No explicit delimiters or "ignore embedded instructions" warnings are used when the agent processes or interpolates paper content.
- Capability inventory: File system manipulation (creating and deleting directories via
shutil), file creation, and execution of Python scripts via the generated sub-skills (uv-check-modelinganduv-check-programming). - Sanitization: No sanitization or validation of the paper content is performed before it is used to generate the exercise pack content.
Audit Metadata