The Agent Skills Directory

[SAFE]: The skill configuration establishes a secure, isolated environment. The frontmatter explicitly defines network: [] and env: [], ensuring that sensitive financial data cannot be exfiltrated to external domains.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external sources that could potentially contain malicious instructions, though the risk is mitigated by the lack of active capabilities.
Ingestion points: Untrusted data is ingested from Qonto via the list_transactions, list_client_invoices, and list_supplier_invoices tools referenced in SKILL.md.
Boundary markers: The skill does not define specific delimiters or instructions to ignore embedded commands within the financial data fields.
Capability inventory: The skill possesses no write tools, no network access, and no system-level execution capabilities; it is limited to reading data and formatting markdown/HTML output.
Sanitization: No explicit sanitization or escaping of external transaction or invoice content is performed before interpolation into the agent's context.
[DATA_EXPOSURE]: The skill includes instructions to protect sensitive information by masking International Bank Account Numbers (IBANs), displaying only the last four digits in any generated reports.

qonto-vat-return