The Agent Skills Directory

[DATA_EXPOSURE]: Authentication is handled via a FREVANA_TOKEN stored in environment variables or passed as an argument. The skill includes instructions to prevent the agent from leaking this token in user responses.
[COMMAND_EXECUTION]: The search_walmart.sh script executes shell commands and Python code to interact with the search API. It performs type checking for integers and numbers and uses whitelist validation for parameters like device types and sort orders to ensure input integrity.
[EXTERNAL_DOWNLOADS]: Product data is retrieved from https://ai-factory.frevana.com. This network operation is consistent with the skill's stated purpose of providing Walmart search results via the Frevana platform.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external product listings which are technically untrusted content.
Ingestion points: Search results from the Frevana API are processed by the script and agent.
Boundary markers: No specific delimiters are defined to isolate untrusted product descriptions in the prompt.
Capability inventory: The skill can write JSON results to the local filesystem in the ./out/ directory.
Sanitization: The script uses Python's json.loads to verify the API response is valid JSON before the agent processes the content.

walmart-search