The Agent Skills Directory

[Indirect Prompt Injection] (HIGH): The skill possesses a significant attack surface for indirect prompt injection due to its file-handling capabilities. Ingestion points: untrusted data enters the agent context via scripts/main.py at line 61 using input_path.read_text(). Boundary markers: absent; the content is read directly without delimiters or instructions to ignore embedded commands. Capability inventory: scripts/main.py contains subprocess-like file system capabilities including output_path.write_text() and output_path.parent.mkdir(). Sanitization: absent; the script performs a direct transformation and write operation without validating or escaping the input content.
[Data Exposure & Exfiltration] (LOW): The script allows the reading of arbitrary file paths passed as command-line arguments. While intended for file processing, this provides the capability to read sensitive system files if the agent is directed to do so by a malicious actor.

skill-name