The Agent Skills Directory

[DATA_EXPOSURE_AND_EXFILTRATION]: The skill processes sensitive financial records (bank statements, ledgers, invoice trackers). It mitigates data exposure risks by computing all matching and triage logic locally. A redact() function is provided and used to mask party names and financial amounts before reconciliation artifacts are shared outside the entitled environment.
[INDIRECT_PROMPT_INJECTION]: The skill ingests external data from various file formats (CSV, XLSX, PDF). This input data is parsed into structured Decimal and date objects, which provides a layer of sanitization. While malicious text in description fields could be presented to the agent in the final report, the deterministic Python engine prevents these strings from being executed as commands during processing.
[DYNAMIC_EXECUTION]: The script uses dynamic path resolution in the _load_engine() function to import shared modules (ingest and dataclean) from the toolkit's root directory. This is a standard architectural pattern for these skills to share a common data-processing core while remaining portable.

data-reconcile