paper-2-web

Warn

Audited by Gen Agent Trust Hub on Jun 30, 2026

Risk Level: MEDIUMREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill instructs the agent to clone an unvetted third-party repository from GitHub (https://github.com/YuhangChen1/Paper2All.git) and execute its Python scripts (pipeline_all.py, pipeline_light.py). Running code from an untrusted source is a high-risk operation that could lead to unauthorized system access or compromise.
  • [EXTERNAL_DOWNLOADS]: The installation process involves fetching external code and dependencies from GitHub and Python package registries.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests untrusted data in the form of academic papers (LaTeX or PDF) to generate structured promotional content.
  • Ingestion points: Paper source files (LaTeX, PDF) provided in the input directories.
  • Boundary markers: No specific delimiters or "ignore embedded instructions" warnings are mentioned to prevent the agent from obeying instructions hidden within the papers.
  • Capability inventory: The skill executes Python scripts, writes files (HTML, JS, PDF, MP4), and makes network calls via the Google Search API.
  • Sanitization: There is no evidence of content validation or sanitization before the paper data is processed by the underlying models.
  • [COMMAND_EXECUTION]: The pipeline relies on system-level command execution to invoke external utilities such as libreoffice and poppler-utils for document format conversion.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jun 30, 2026, 03:46 PM
Security Audit — agent-trust-hub — paper-2-web