The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external VTEX commerce APIs (catalog, search, and checkout). This untrusted data is processed and presented to the agent, creating a potential vector for malicious content to influence agent logic.
Ingestion points: Data is fetched through the client.ts functions (vtexFetch, vtexCachedFetch, intelligentSearch) and various files in the loaders/ directory.
Boundary markers: The skill uses utils/transform.ts to map raw VTEX responses into schema.org formats, providing a degree of data normalization.
Capability inventory: The skill manages network requests to commerce endpoints and handles browser cookies for session and cart state.
Sanitization: While transformations are applied, there is no explicit evidence of sanitization specifically designed to neutralize prompt injection within the ingested data.
[COMMAND_EXECUTION]: The skill documentation describes the use of build-time code generation tools within the developer's local environment.
Evidence: References like references/cart-wiring.md and references/audit-checklist.md mention running bun run generate and using bunx tsx to execute the @decocms/blocks-cli tool for generating server-side function bindings.

apps-vtex