ai-co-scientist
Pass
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- COMMAND_EXECUTION (LOW): The skill automates scientific experimentation by writing and executing Python scripts. While this involves dynamic execution, it is the primary intended function of the skill. Execution is managed through the documented
tree.pyscript.\n- EXTERNAL_DOWNLOADS (SAFE): The visualization template loads the p5.js library from a public CDN, which is a standard method for including web dependencies.\n - Evidence:
assets/viz-template/template.htmlloadsp5.min.jsfromcdnjs.cloudflare.com.\n- PROMPT_INJECTION (LOW): Thescripts/visualize.pyscript generates an HTML report by injecting raw experiment data (including terminal output and analysis) into a template. This creates an indirect prompt injection surface where malicious experiment output could execute scripts in the user's browser (XSS).\n - Ingestion points:
scripts/visualize.pyreads data from project tree files (e.g.,stage_3_iter_1.json).\n - Boundary markers: None present for the visualization data interpolation.\n
- Capability inventory: The skill can execute local Python code and generate HTML reports.\n
- Sanitization: Absent; data is placed into the HTML template without escaping for script context.
Audit Metadata