The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted data from images and PDFs (invoices) and interpolates extracted values directly into a shell command script.
Ingestion points: Reads content from user-provided invoice images and PDFs via vision/read tools.
Boundary markers: Absent. No specific delimiters or warnings are used to isolate the data extracted from invoices.
Capability inventory: Includes file system operations (find, rm, zip, unzip, mv) and arbitrary code execution via python3 subprocess calls.
Sanitization: Absent. The instructions do not mandate sanitizing or escaping the extracted invoice data (e.g., amount strings) before inserting them into the Python validation script template.
[COMMAND_EXECUTION]: The skill performs extensive shell command execution to automate its workflow.
Generates and executes a temporary Python script using python3 -c to perform arithmetic validation on extracted totals.
Uses find with the -delete flag to remove specific file types (.xml, .ofd, .zip) from the user's directory.
Uses zip, unzip, and mv for file management and archive processing.
Employs rm -rf and rm -f for cleaning up temporary directories and original files.

wlzh-invoice-scanner