The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements logic to ingest and process untrusted data from external sources, which could be exploited to influence the agent's behavior via instructions hidden in the processed data.\n
Ingestion points: Data enters the context through the extract_invoice_data method (which uses OCR to process invoice images/PDFs in the InvoiceProcessingBot class) and the get_erp_orders method (which fetches JSON payloads from external ERP APIs in the SupplyChainAPIIntegration class).\n
Boundary markers: The instructions and provided code blocks lack explicit boundary markers or directives that would instruct the model to treat the incoming external data as non-executable content.\n
Capability inventory: The skill possesses significant capabilities including browser automation via selenium, network communication via requests, and email transmission via smtplib, creating a functional path for malicious instructions to trigger external actions.\n
Sanitization: Content extracted from OCR or API payloads is processed directly via regex or simple field extraction without comprehensive sanitization, validation, or filtering before being used to populate fields in web forms or API calls.

supply-chain-automation