academic-figure-generation

Warn

Audited by Gen Agent Trust Hub on May 4, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill instructs the user to clone a third-party repository from GitHub (https://github.com/dwzhu-pku/PaperBanana) as part of the setup process.
  • [REMOTE_CODE_EXECUTION]: The scripts/generate.py script dynamically adds the path of the cloned repository to the Python system path (sys.path.insert(0, str(project_root))) and imports several modules (e.g., agents.planner_agent, agents.visualizer_agent). This results in the execution of code from an external, unverified source.
  • [COMMAND_EXECUTION]: The documentation provides shell commands for cloning repositories, installing dependencies using uv, and executing the generation script with various parameters.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted paper text provided by the user.
  • Ingestion points: The scripts/generate.py script reads the full content of a file provided via the --method-file argument.
  • Boundary markers: The script does not utilize delimiters or specific instructions to isolate the ingested text from the agent's logic.
  • Capability inventory: The skill has the capability to write files (PNG images and results.json) and triggers network activity through the underlying LLM and image generation models.
  • Sanitization: There is no evidence of sanitization or validation of the input text before it is passed to the multi-agent pipeline.
Audit Metadata
Risk Level
MEDIUM
Analyzed
May 4, 2026, 11:18 AM
Security Audit — agent-trust-hub — academic-figure-generation