The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill contains various scripts (e.g., check_deidentification.py, extract_clinical_data.py, validate_case_report.py) that are intended to parse and analyze external clinical reports, which constitutes a potential ingestion point for malicious instructions.
Ingestion points: Clinical report files provided to the validation and data extraction scripts via command-line arguments.
Boundary markers: The skill documentation emphasizes privacy and accuracy but does not define strict technical boundaries or delimiters to isolate processed clinical text from the agent's control flow.
Capability inventory: The skill utilizes the Bash tool to execute Python scripts that can read files, write JSON results, and copy template files (shutil.copy).
Sanitization: The scripts process external data using regular expressions for detection and extraction purposes; however, they do not specifically sanitize or escape clinical text to prevent it from being interpreted as instructions by an agent during analysis.

clinical-reports