The Agent Skills Directory

[SAFE]: The core logic is implemented in a standalone Python script (deidentify.py) that uses standard libraries and the openpyxl package for file processing. Analysis confirms the script performs no network operations, and its file system interactions are limited to reading input data and writing de-identified outputs, reports, and logs.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted clinical data provided by the user in CSV, TSV, or XLSX formats. Maliciously crafted data could theoretically attempt to influence the agent via the scan reports or audit logs that the agent is permitted to read.
Ingestion points: The load_tabular function in deidentify.py reads user-supplied data files.
Boundary markers: The SKILL.md file provides clear instructions and safety rules for the agent, explicitly forbidding it from reading raw data or mapping files.
Capability inventory: The skill has the capability to read and write files and execute local Python code via the Bash tool.
Sanitization: The script uses robust regex-based detection to redact or pseudonymize PHI before any summaries are presented to the agent, reducing the risk of malicious data influencing the agent's behavior.
[EXTERNAL_DOWNLOADS]: The SKILL.md documentation advises the user to install the openpyxl library via pip to enable Excel file support. This is a well-known, legitimate package from an official package registry commonly used for handling Excel data.

deidentify