skills/edwinhu/workflows/workshop/Gen Agent Trust Hub

workshop

Pass

Audited by Gen Agent Trust Hub on Sep 7, 2026

Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The core logic in scripts/workshop-deck.py and scripts/checks/check-overflow.sh uses shell execution and subprocess.run to call the Typst compiler and Python-based slide analysis tools. These operations are essential for generating and validating the presentation artifacts.
  • [DYNAMIC_EXECUTION]: The script references/constraints/run-constraints.py utilizes importlib.util to dynamically load local Python modules (typst-*.py) for slide formatting and convention checks. This allows for modular verification logic while remaining restricted to the skill's own local directory.
  • [EXTERNAL_DOWNLOADS]: The skill fetches the pypdf, lxml, and pytest packages from the Python Package Index (PyPI) using uv run. PyPI is a well-known service, and these dependencies are required for document parsing and testing.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted research papers during the presentation generation process. It provides strong mitigation against injection attacks through a 'computed deck gate' that strictly verifies all generated artifacts against an approved plan using machine-parsed grammars and independent validation agents.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 7, 2026, 05:55 AM
Security Audit — agent-trust-hub — workshop