The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external financial institutions, which introduces a potential surface for indirect prompt injection through unvalidated transaction strings.
Ingestion points: The skill fetches Transaction objects containing string fields such as transactionDescription, merchantName, and originalTransactionDescription via the FinanceStore API, as documented in references/financekit-patterns.md.
Boundary markers: The implementation examples and instructions do not include boundary delimiters or explicit warnings for the agent to ignore instructions potentially embedded within these transaction fields.
Capability inventory: The skill provides instructions for reading sensitive financial records (accounts, balances, and transactions) and persisting this data using UserDefaults and App Group shared containers, as seen in references/financekit-sync-and-lifecycle.md.
Sanitization: No evidence of validation, filtering, or sanitization for merchant-controlled text fields is provided in the skill's code snippets or patterns.

financekit