The Agent Skills Directory

[Data Exposure] (HIGH): Path Traversal in functions/command_validator.py. The validate_command_file function reads file contents from a user-provided path using Path(file_path).read_text() without validation or directory restriction, enabling access to arbitrary system files (e.g., ~/.ssh/id_rsa).
[Indirect Prompt Injection] (LOW): Surface for indirect prompt injection in functions/command_generator.py.
Ingestion points: Untrusted strings in the description and sections arguments of generate_command in functions/command_generator.py.
Boundary markers: Absent. No delimiters are used to isolate user-provided content from the markdown instruction templates.
Capability inventory: The generated markdown files serve as the primary instruction set for the Navigator agent.
Sanitization: None; the script interpolates raw input directly into f-strings, allowing malicious instructions to be embedded in generated documentation.

plugin-slash-command