The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The Python implementation example provided in references/clients.md specifies the use of a library called httpx2. This is not the standard community-supported library (which is httpx), and such naming conventions are frequently associated with typosquatting attacks where malicious code is distributed under names similar to popular packages. Encouraging users to install and run code from non-verified dependencies represents a risk of remote code execution.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to fetch and process data from external, untrusted sources including Amazon, Walmart, and eBay. This scraped content is then directly fed into the agent's context. The lack of strict boundary markers or sanitization for adversarial instructions within product metadata (such as titles or descriptions) creates a surface for indirect prompt injection.
Ingestion points: The skill retrieves product details using the get-dataset-items tool within the logic defined in SKILL.md and references/clients.md.
Boundary markers: While SKILL.md suggests prompt instructions regarding data freshness, it does not include explicit delimiters to isolate external product data from the system instructions.
Capability inventory: The skill utilizes tools to execute Apify Actors (apify--e-commerce-scraping-tool) and access datasets, providing the agent with the ability to perform complex scraping operations.
Sanitization: The references/fields.md document outlines methods for cleaning product names and parsing prices, but these are primarily for data normalization rather than security-focused sanitization against malicious prompts.
[EXTERNAL_DOWNLOADS]: The skill connects to a remote MCP server at https://mcp.apify.com and references a external GitHub repository https://github.com/luispintoapify/ecommerce-agent-starter. These endpoints are controlled by the skill's vendor and are necessary for the tool's integration with the Apify platform.

apify-product-data-setup