The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from multiple sources, including image text (via OCR) and CSV mapping files for renaming operations. A maliciously crafted CSV could be used to manipulate file system operations. While the script requires user confirmation for renames, an automated agent might bypass this check.
[COMMAND_EXECUTION]: The cmd_rename function in scripts/batch_pipeline.py is vulnerable to path traversal. It uses os.path.join to construct source and destination paths from CSV input without sanitizing for directory traversal characters (e.g., ../). This allows an attacker to move files to arbitrary locations outside the target directory.
[DYNAMIC_EXECUTION]: The cmd_verify function accepts a user-provided regular expression and applies it to filenames. A complex or malicious regex (ReDoS) could lead to high CPU consumption and denial of service.
[DATA_EXPOSURE]: The cmd_dedupe function reads entire files into memory to calculate SHA-256 hashes (open(f, "rb").read()). Processing extremely large files could lead to memory exhaustion (Out-of-Memory) and system instability.

batch-file-pipeline