The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The script scripts/unmix_repomix.py possesses a directory traversal vulnerability surface. It extracts file paths from untrusted input files and uses them to write files to the filesystem without validation. If an agent processes a malicious repomix file containing absolute paths or traversal sequences, it could overwrite sensitive system files.\n- Ingestion points: Data enters the skill via the repomix_file processed by scripts/unmix_repomix.py.\n- Boundary markers: None. There are no instructions for the agent to verify the contents of the repomix file or to use protective delimiters to prevent the execution of embedded instructions.\n- Capability inventory: The skill includes a Python script that can create directories and write files to arbitrary paths if not properly sanitized.\n- Sanitization: Absent. The functions unmix_xml, unmix_markdown, and unmix_json directly use the file_path from the input file to create a Path object without checking if it escapes the intended output_dir.

repomix-unmixer