The Agent Skills Directory

[SAFE]: The skill relies on authenticated vendor scripts (@reduck/amazon.com/*) for all interactions with Amazon, ensuring a trust-based interaction within the provider's ecosystem.
[SAFE]: Documents the use of a local bridge (reduck local --cookies) to handle authenticated data retrieval for product reviews, following established platform security practices.
[SAFE]: Implements regex validation for input URLs in the category list function to ensure only expected Amazon search pages are processed.
[PROMPT_INJECTION]: The skill processes untrusted customer reviews and product listings from Amazon, presenting an indirect prompt injection surface. This is evaluated as safe given the skill's constrained read-only scope.
Ingestion points: Aggregated product listings and star ratings fetched via search-products, get-product, and get-product-reviews.
Boundary markers: Absent; the system prompt does not specify delimiters for external content.
Capability inventory: Interaction is strictly limited to execution of vendor-specific tools via run_script for data retrieval; no system-level or destructive capabilities are present.
Sanitization: No content filtering or validation of the retrieved text is described.

amazon-research