The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external medication data in JSON format, which represents a potential injection surface if the data source is untrusted.
Ingestion points: The script scripts/main.py reads patient medication lists from user-provided file paths via the --pre-admission and --inpatient arguments.
Boundary markers: The workflow in SKILL.md requires a mandatory PHI (Protected Health Information) confirmation step before processing. The output uses structured JSON and fixed response templates to delimit data.
Capability inventory: The skill is limited to local file system read/write for JSON reports and standard output. It has no network capabilities or arbitrary system command execution.
Sanitization: The script uses structured dataclasses (Medication) and performs name normalization (lowercase, space removal), which provides basic structural validation of the input data.
[COMMAND_EXECUTION]: The SKILL.md documentation includes instructions for running syntax checks (python -m py_compile) and CLI help commands. These are standard development practices for verifying the local environment and do not involve untrusted data flow into shell execution.

medication-reconciliation