Skills
Navigate
skills/a-green-hand-jack/icml-skills/icml-camera-ready/Gen Agent Trust Hub

icml-camera-ready

Pass

Audited by Gen Agent Trust Hub on Oct 5, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The script scripts/check_submission.py utilizes subprocess.run to call local system utilities such as pdfinfo and pdftotext. These calls are implemented using list-based arguments rather than shell strings, preventing shell injection vulnerabilities.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted LaTeX and PDF files provided by the user.
  • Ingestion points: Paper source and PDF files are loaded in scripts/check_submission.py via load_tex and check_pdf functions.
  • Boundary markers: The skill does not explicitly define markers when passing content to the agent context.
  • Capability inventory: Scripts perform file system operations (read/write) and execute local command-line tools for PDF analysis.
  • Sanitization: The skill includes a proactive INJECTION-PHRASE detector within scripts/check_submission.py designed to identify and flag adversarial prompts hidden in academic paper text.
  • [SAFE]: The skill documentation encourages security best practices, such as explicitly instructing the user to remove internal notes and credentials from the workspace before publishing research code.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 5, 2026, 12:10 PM