The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from Amazon product listings, including titles, bullet points, and raw HTML content. This creates an attack surface where a malicious seller could attempt to influence the agent's behavior.
Ingestion points: Data is retrieved via the LinkFox API and extracted from raw HTML files specified in the products[].rawHtmlFile field.
Boundary markers: The instructions (SKILL.md) explicitly direct the agent to distinguish between different data fields (e.g., Title vs. Item Highlights) and to treat HTML as raw data without executing scripts.
Capability inventory: The skill utilizes Python scripts (amazon_product_detail.py) to perform network requests and local file I/O within the project directory.
Sanitization: Instructions require the agent to decode HTML entities and organize whitespace without generating or rewriting the original product copy.
[COMMAND_EXECUTION]: The skill relies on local Python scripts (scripts/amazon_product_detail.py and scripts/onboarding.py) to interact with the vendor's API gateway and manage user authentication. These scripts are invoked via standard subprocess calls.
[PERSISTENCE]: Documentation in references/onboarding.md provides instructions for users to manually add API keys to their shell profiles (e.g., .bashrc, .zshrc) to maintain authentication across sessions. This is a standard configuration practice for CLI-based developer tools and is presented as a guided setup rather than an automated hidden action.
[DATA_EXPOSURE]: The skill communicates with vendor-controlled domains (linkfox.com and its subdomains) to perform its core functions. It uses environment variables for sensitive API keys and implements local caching to minimize redundant network requests.

linkfox-amazon-product-detail