The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from Amazon product pages and order history, which could contain malicious instructions designed to influence the agent's behavior.
Ingestion points: The agent reads content from Amazon search results, product pages, and order history pages as part of the ordering and return workflows defined in SKILL.md.
Boundary markers: None. The instructions do not include delimiters or specific commands to ignore instructions embedded within the webpage content.
Capability inventory: The agent has the capability to navigate URLs, click UI elements (including "Buy it again" and "Place order"), and read environment variables containing sensitive user preferences (AMAZON_PAYMENT_METHOD, AMAZON_SHIPPING_ADDRESS).
Sanitization: None. Data extracted from the browser session is processed directly to identify items and interact with the Amazon store interface.
[COMMAND_EXECUTION]: The skill provides shell commands to launch Google Chrome with remote debugging enabled and to connect the agent-browser CLI tool.
The commands utilize the --remote-debugging-port flag, which exposes the browser session to control by the agent or other processes.
The skill instructs the agent to use a persistent browser profile located at $HOME/.config/chrome-agent, which stores sensitive session cookies and login state for the Amazon account.

amazon