data-science-visualization
Warn
Audited by Gen Agent Trust Hub on Feb 16, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- COMMAND_EXECUTION (MEDIUM): The skill provides numerous instructions for executing shell commands and running local servers. Specific examples include 'bokeh serve app.py' in 'references/bokeh-server.md', 'jupyter nbconvert' and 'voila' in 'references/sharing-publishing.md', and 'pytest --nbval' in 'references/notebook-testing.md'. These tools execute local code or convert code to other formats, which can be exploited if the source notebooks are untrusted.\n- REMOTE_CODE_EXECUTION (MEDIUM): In 'references/plotly-dash.md', the provided template includes 'app.run(debug=True)'. The Dash/Flask debugger is an interactive tool that allows arbitrary code execution from the browser. If an agent executes this code and the port is accessible to others, it creates a direct path for RCE.\n- EXTERNAL_DOWNLOADS (LOW): The skill documentation suggests installing several third-party packages via pip, such as 'nbval', 'voila', 'streamlit-aggrid', and 'streamlit-echarts'. While these are standard packages in the data science ecosystem, they represent external dependencies that are not pinned to specific versions.\n- INDIRECT_PROMPT_INJECTION (MEDIUM): The skill's purpose is to process and visualize data. The data ingestion point (typically Pandas or Polars dataframes) is an attack surface. Malicious strings embedded in the data could potentially exploit XSS vulnerabilities in the interactive components of Plotly, Bokeh, or Dash dashboards when rendered in a browser.
Audit Metadata