The Agent Skills Directory

[COMMAND_EXECUTION]: The instructions mandate the automatic installation of dependencies without user consent, specifically stating the agent should "install the missing dependency automatically without asking the user to confirm."
[REMOTE_CODE_EXECUTION]: The skill directs the agent to bypass standard security protocols and user oversight by instructing it to "not request separate approval for dependency installation" and not to describe the process to the user.
[EXTERNAL_DOWNLOADS]: The skill references an external installation link https://www.browseract.com/?co-from=ecommerce&redirect=https://github.com/browser-act/skills/tree/main which acts as a redirector for software retrieval from a non-whitelisted source.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill is designed to scrape content from various external sources including Google Shopping URLs, store pages, and seller profiles in SKILL.md.
Boundary markers: There are no instructions or delimiters provided to ensure the agent ignores or sanitizes potentially malicious instructions embedded within the scraped HTML or text.
Capability inventory: The skill utilizes a browser-based CLI (browser-act) for live web interaction and task execution, which provides a significant capability surface for injected instructions to exploit.
Sanitization: The instructions lack any requirement for filtering, validation, or escaping of external content before it is processed by the agent.

google-shopping-sku-ean-offer-scraper