clinical-reports
Pass
Audited by Gen Agent Trust Hub on Apr 22, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: Attack surface for indirect prompt injection. The skill ingests untrusted clinical data through validation and extraction scripts such as extract_clinical_data.py, validate_case_report.py, and check_deidentification.py.
- Ingestion points: Clinical report files (Markdown or text) provided for validation or data extraction.
- Boundary markers: No delimiters or warnings are used to differentiate clinical data from instructions during processing.
- Capability inventory: The skill possesses Bash, Write, and Edit permissions, which allows for script execution and file system modification.
- Sanitization: Input data is processed using regular expressions for extraction, but there is no mechanism to sanitize or ignore embedded natural language instructions within the medical text.
- [SAFE]: The skill's utility scripts for HIPAA compliance and terminology validation (e.g., compliance_checker.py, terminology_validator.py) use standard library modules and perform legitimate medical documentation tasks. No evidence of malicious code, data exfiltration, or unauthorized network activity was found in the provided file set.
Audit Metadata