The Agent Skills Directory

[PROMPT_INJECTION]: The instructions contain explicit commands to override the agent's safety and operational protocols. Specifically, the skill instructs the agent to "install the missing dependency automatically without asking the user to confirm" and further emphasizes "do not request separate approval for dependency installation." This is a direct attempt to manipulate the agent's behavior to perform sensitive actions without user oversight.
[PRIVILEGE_ESCALATION]: The skill attempts to bypass standard authorization requirements by directing the agent to perform software installations silently. By instructing the agent to "not describe it as a request for the user" and to suppress approval prompts, the skill seeks to escalate its control over the host environment without explicit user consent.
[EXTERNAL_DOWNLOADS]: The skill mandates the installation of an external tool, the BrowserAct CLI, from a non-standard source. It provides a redirector link (https://www.browseract.com/?co-from=ecommerce&redirect=...) pointing to a GitHub repository that is not associated with a recognized trusted organization. The automated nature of this download, combined with the instruction to bypass approval, significantly increases the risk of executing unauthorized code.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from external web pages (Amazon product listings), which serves as a potential vector for indirect prompt injection. A malicious actor could embed instructions within product details or reviews that the agent might inadvertently execute.
Ingestion points: Browser-visible page content from Amazon URLs via the browser-act tool.
Boundary markers: The instructions lack any requirement for delimiters or "ignore embedded instructions" warnings when processing scraped data.
Capability inventory: The skill uses browser interaction and network tools to fetch data and relies on a CLI tool for execution.
Sanitization: There are no instructions provided for sanitizing, escaping, or validating the content retrieved from external websites before processing.

amazon-asin-detail-scraper