The Agent Skills Directory

[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection due to its core function of reading and analyzing untrusted external files.
Ingestion points: The scripts scripts/config_validator.py, scripts/hipaa_scan.py, and scripts/report_generator.py are designed to recursively scan all files in a target directory.
Boundary markers: There are no delimiters or instructions in the SKILL.md or the Python scripts to differentiate between the agent's instructions and the content being analyzed, nor are there commands to ignore embedded instructions.
Capability inventory: The skill is granted extensive permissions, including Read, Write, Edit, and Bash command execution, which could be leveraged if an injection occurs.
Sanitization: The skill does not implement any mechanisms to sanitize or filter the content of the files it reads before providing that data to the agent's context.

checking-hipaa-compliance