The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function involves processing external clinical reports and data files (CSV, Markdown) which are untrusted ingestion points.
Ingestion points: Files are read by validate_case_report.py, check_deidentification.py, extract_clinical_data.py, and format_adverse_events.py for validation and data extraction.
Boundary markers: The provided scripts do not implement specific boundary markers or delimiters to differentiate between clinical data and potential adversarial instructions within the reports.
Capability inventory: The skill is granted Bash, Write, and Edit permissions, which provide a significant capability surface for any instructions potentially processed from the input data.
Sanitization: Input text is processed using regular expressions for pattern matching, but no sanitization or filtering of instruction-like strings is performed prior to the agent's contextual processing of the script outputs.

clinical-reports