The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface for indirect prompt injection because it processes transaction data from external sources without explicit sanitization or boundary markers.
Ingestion points: The skill ingests untrusted data through the transactions_new, transactions_history, and accounts inputs, specifically within fields like merchant and description_raw where natural language instructions could be embedded.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to distinguish between data and instructions or to ignore commands hidden within the transaction fields.
Capability inventory: The skill's primary capability is to generate a structured JSON report. It also mentions appending alert logs to reports/alerts/, representing a file-write operation. No subprocess calls, network operations, or dynamic code execution capabilities were identified.
Sanitization: The skill does not implement validation, escaping, or filtering of the external transaction data before it is processed by the agent logic.

anomaly-fraud-scanner