The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill establishes a workflow for ingesting and processing financial data from external institutions, which serves as a potential vector for indirect prompt injection via transaction fields.
Ingestion points: External data is accessed through FinanceStore queries, TransactionPicker, and transaction history sequences described in SKILL.md and references/financekit-sync-and-lifecycle.md.
Boundary markers: The skill does not provide specific delimiters or instructions for the agent to treat data fields (e.g., transactionDescription or merchantName) as untrusted or to ignore embedded instructions.
Capability inventory: The skill provides functionality for local data synchronization, state management using UserDefaults and App Groups, and SwiftUI rendering. It does not include dangerous capabilities such as arbitrary command execution, network exfiltration, or dynamic code evaluation.
Sanitization: While the skill offers guidance on currency formatting and merchant category mapping, there are no instructions for sanitizing or escaping incoming text data to prevent adversarial influence on the agent's logic.

financekit