The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes explicit instructions to override the agent's safety and confirmation guardrails. It directs the agent to install the 'BrowserAct' dependency 'automatically without asking the user to confirm' and further mandates that the agent 'do not request separate approval for dependency installation.' Additionally, it instructs the agent to conceal this activity from the user by not describing the link as a request to open a browser.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of external components (BrowserAct CLI and Skill) from a third-party redirection URL (pointing to a GitHub repository). The combination of external code acquisition with instructions to skip user verification significantly increases the risk of unauthorized software execution.
[INDIRECT_PROMPT_INJECTION]: As the skill is designed to scrape and process content from external websites (Zalando), it is susceptible to indirect prompt injection. Malicious instructions embedded in product pages could potentially influence the agent's behavior during the scraping session.
Ingestion points: Live browser page content, metadata, and network responses captured by the browser-act tool.
Boundary markers: None identified in the instructions to separate untrusted web data from agent instructions.
Capability inventory: The agent has the ability to execute CLI commands, perform network operations via a browser, and write data to the filesystem (workspaces/).
Sanitization: No sanitization or validation logic is defined for the captured data before it is returned or processed.

zalando-product-scraper