The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data (such as product descriptions and customer reviews) retrieved from external marketplaces like Amazon, eBay, and Walmart. This data is ingested into the agent's context, creating a potential surface for instructions hidden within marketplace listings to influence agent behavior.
Ingestion points: Untrusted content enters the agent's context through JSON responses from various search and product-detail endpoints called via scripts/crawlora.sh.
Boundary markers: The skill does not provide explicit delimiters or "ignore embedded instructions" warnings for the processed data.
Capability inventory: The skill's capabilities are limited to making network requests to a specific vendor API; it lacks broad file system access or generic system command execution.
Sanitization: The helper script strictly sanitizes request parameters, though the agent remains responsible for safely handling the resulting JSON data.
[COMMAND_EXECUTION]: The skill utilizes a shell helper script (scripts/crawlora.sh) to perform API requests. The script is highly defensive, implementing a comprehensive whitelist for HTTP methods and API routes. It prevents command injection by avoiding shell interpolation and safeguards against local file disclosure by rejecting the @ character in query parameters and streaming the POST body to curl via standard input.

product-price-research