The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data files which could contain malicious instructions designed to influence the agent's behavior at runtime.
Ingestion Points: The skill ingests data from Hashavshevet legacy files (HESHIN.dat, PKUDOT.dat, MANOT.dat, KARTIS.dat, HESHBON.dat, MATZAV.dat, TNUOT.dat), uniform structure files (INI.TXT, BKMVDATA.TXT), and bank transaction CSVs (e.g., Leumi CSV).
Boundary Markers: The provided Python scripts and instructions do not implement explicit boundary markers or warnings to the agent to disregard natural language instructions found within the data records.
Capability Inventory: The skill uses Bash(python:*), Read, Edit, and Write tools to parse, convert, and export data, providing a vector for instructions within the data to trigger unwanted actions if interpreted as commands.
Sanitization: While the code uses errors='replace' and errors='strict' for character encoding integrity, there is no validation or filtering of the actual text content (e.g., description fields) to prevent prompt injection.
[COMMAND_EXECUTION]: The skill uses Python execution via the Bash tool to perform data transformations and validations.
Evidence: Code snippets for parse_fixed_width, detect_and_convert, and generate_hashavshevet_import are intended to be executed by the agent to manipulate file content and perform accounting logic.

hashavshevet-data-tools