The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructions involve the agent reading product information (names, HS codes, and descriptions) from a database to generate customs declarations and duty estimates. This creates a surface for indirect prompt injection if the database contains adversarial content. • Ingestion points: Database calls in customs-declaration.ts and input arguments in screen-order.ts and duties-estimation.ts (SKILL.md, task.md). • Boundary markers: No delimiters or specific instructions to ignore embedded commands within the product data are provided. • Capability inventory: The skill generates code for database migrations and constructs payloads for network operations with external APIs (EasyPost). • Sanitization: There is no mention of sanitizing or validating product-derived strings before they are used in code generation or API payloads.
[SAFE]: The skill instructions specify that sensitive information, such as API keys (EASYPOST_API_KEY) and signer names (CUSTOMS_SIGNER_NAME), should be retrieved from environment variables rather than being hardcoded in the scripts, which aligns with security best practices.

international-shipping