clinical-decision-support

Pass

Audited by Gen Agent Trust Hub on Apr 12, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is designed to process external patient and clinical data to generate reports, creating an attack surface for indirect prompt injection.\n
  • Ingestion points: Untrusted data enters the agent's context through CSV, JSON, and text files processed by scripts such as biomarker_classifier.py, generate_survival_analysis.py, and build_decision_tree.py.\n
  • Boundary markers: Absent. The skill instructions do not specify the use of delimiters or warnings to the agent to ignore potentially malicious instructions embedded within the clinical data.\n
  • Capability inventory: The skill has access to the Bash, Write, Edit, and Read tools, providing a broad set of capabilities that could be exploited if an injection occurs.\n
  • Sanitization: While the skill includes a validate_cds_document.py script, it focuses on medical quality and HIPAA compliance rather than sanitizing input against code or prompt injection.\n- [COMMAND_EXECUTION]: The SKILL.md file instructs the agent to execute a shell command (python scripts/generate_schematic.py) using natural language descriptions. If these descriptions are derived from untrusted user input without proper escaping, it presents a risk of command injection.\n- [COMMAND_EXECUTION]: The script build_decision_tree.py dynamically assembles LaTeX/TikZ code by interpolating user-provided strings. This creates a vulnerability to LaTeX injection, which could be used to read files or execute unintended logic if the resulting document is compiled in a privileged environment.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 12, 2026, 08:27 AM