The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes shell commands including find, unzip, mv, rm, and zip to perform file operations, cleanup, and archival tasks within user-specified directories.
[DYNAMIC_EXECUTION]: The skill generates and executes a Python command string at runtime using the python3 -c flag. This command is constructed by interpolating data extracted from scanned invoices into a script template to perform arithmetic validation. This represents a dynamic code execution risk if malicious data is extracted from an invoice and interpreted as code.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external files (images, PDFs, and ZIP archives) from the file system. These files serve as the primary input for the agent's data extraction tasks and could potentially contain malicious content intended to influence the agent's output or actions.
Ingestion points: Files identified and read via the find command and the Read tool within the user-provided directory in SKILL.md (Workflow Step 2).
Boundary markers: None identified; the skill does not explicitly instruct the agent to ignore embedded instructions within the processed invoice data.
Capability inventory: Access to shell command execution (find, unzip, mv, rm, zip) and the ability to execute generated Python scripts (python3 -c).
Sanitization: No sanitization or validation of extracted text is described before it is used in the Python validation script or the final report.

wlzh-invoice-scanner