The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources and transaction records which could potentially contain instructions aimed at influencing the agent's behavior.
Ingestion points: Data is ingested from transaction counterparty names and descriptions via list_transactions, client names and notes via list_client_invoices, and external dataset metadata/content via search_datasets.
Boundary markers: The instructions do not specify the use of delimiters or 'ignore embedded instructions' prompts when processing transaction labels or dataset strings.
Capability inventory: The skill has access to sensitive financial organizational data and the ability to generate files using the Write tool.
Sanitization: The skill implements sensitive data masking for IBANs but does not explicitly mention sanitization or escaping for free-text fields like transaction descriptions or invoice notes before they are interpolated into reports.

qonto-sector-benchmark