The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains explicit instructions to override safety and transparency protocols regarding user consent. It commands the agent to install dependencies "automatically without asking the user to confirm" and explicitly forbids the agent from describing the action to the user or requesting "separate approval for dependency installation."
[REMOTE_CODE_EXECUTION]: The skill directs the agent to download and install a third-party command-line interface (browser-act CLI) from a remote URL (https://www.browseract.com/...) if it is missing. Performing unverified software installation without human intervention is a high-severity risk for executing malicious code.
[EXTERNAL_DOWNLOADS]: The skill orchestrates the download of executable tools from a non-trusted external domain (browseract.com) and GitHub organization (browser-act) while deliberately concealing the activity from the user.
[PROMPT_INJECTION]: The skill possesses a vulnerability surface for indirect prompt injection by processing untrusted data from eBay.
Ingestion points: The skill ingests marketplace content including seller profiles, catalogs, and "browser-visible network responses" via the browser-act tool.
Boundary markers: There are no instructions to use delimiters or ignore embedded instructions within the scraped web content.
Capability inventory: The skill utilizes the browser-act CLI for browser interaction and network data capture, and it has the capability to write results to the workspace.
Sanitization: No sanitization or validation logic is defined for the content retrieved from external pages before it is processed by the AI model.

ebay-seller-profile-scraper