workbook
Pass
Audited by Gen Agent Trust Hub on Oct 26, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied content such as math questions from PDFs or lists and interpolates them directly into LaTeX files (
body.tex) for compilation. - Ingestion points: User-supplied questions (PDF, screenshots, or list) in Step 1.
- Boundary markers: None specified for the interpolation of question text into LaTeX.
- Capability inventory: Subprocess calls to
tectonic(LaTeX compiler),pdftoppm(image processor), and SymPy script execution in the scratchpad. File system writes tobody.texandrevision.tex. - Sanitization: No sanitization or escaping of LaTeX special characters or commands is mentioned, which could allow malicious input to interfere with the build process.
- [COMMAND_EXECUTION]: The skill executes external CLI tools to build and preview the workbook.
- Evidence: Uses
tectonic -X compile revision.texfor document compilation andpdftoppm -r 70 -pngfor image generation. - [EXTERNAL_DOWNLOADS]: The skill contains instructions to download and install a binary if the required compiler is missing.
- Evidence: Step 5 instructs the agent to install the
tectonicofficial release binary into~/.local/opt/tectonicif it is not found on the system. - [DYNAMIC_EXECUTION]: The skill generates and runs a Python script to verify mathematical results.
- Evidence: Step 2 involves writing a SymPy script in the scratchpad that computes all final answers to compare against official solutions.
Audit Metadata