rct-bias-assessment-rob

Pass

Audited by Gen Agent Trust Hub on Aug 1, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill provides a structured workflow for medical paper analysis. Its functionality is implemented via two Python scripts: scripts/assess_rob2.py and scripts/extract_pdf.py.- [EXTERNAL_DOWNLOADS]: The skill uses the PyPDF2 Python library for text extraction. This is a standard, well-known library for processing PDF documents.- [INDIRECT_PROMPT_INJECTION]: The skill processes external research data (RCT papers) to extract specific bias markers. While this creates a surface for indirect prompt injection (e.g., a paper containing instructions to bias the assessment), the skill's narrow focus on specific ROB2 domains and lack of high-privilege capabilities (like network access or system file modification) limits the potential impact to incorrect analysis results. No evidence of malicious exploitation was found.- [DATA_EXFILTRATION]: The skill includes a script to write extracted text to a local file. Analysis of the code shows no network operations, use of sensitive system paths, or credential harvesting.- [SAFE]: The skill includes a metadata file (rct-bias-assessment-rob2_audit_result_v1.json) which contains historical evaluation data and assertion results. This file does not contain executable code or malicious instructions.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 1, 2026, 10:43 AM
Security Audit — agent-trust-hub — rct-bias-assessment-rob