The Agent Skills Directory

[DATA_EXFILTRATION]: The skill facilitates connections to remote endpoints such as ubc.oxylabs.io and ubs.oxylabs.io. These are official service endpoints for the skill's author, Oxylabs. The documentation correctly advises users to handle credentials using environment variables like OXY_UNBLOCKER_USERNAME to prevent sensitive data exposure in source code.- [INDIRECT_PROMPT_INJECTION]: The skill exposes an attack surface for indirect prompt injection by processing untrusted data from the web. 1. Ingestion points: Data enters the agent's context through methods like page.goto(), page.content(), and page.$$eval() as shown in SKILL.md and examples.md. 2. Boundary markers: The provided code examples do not include explicit delimiters or instructions to ignore embedded content within processed web pages. 3. Capability inventory: The skill uses Playwright and Puppeteer, which provide capabilities for browser automation, JavaScript execution via page.evaluate(), and local file writing for screenshots and PDFs (page.screenshot, page.pdf). 4. Sanitization: There is no evidence of sanitization or filtering of the HTML content retrieved from remote sites.- [REMOTE_CODE_EXECUTION]: The skill enables remote execution of browser logic via the Chrome DevTools Protocol (CDP) on the vendor's infrastructure. This is standard behavior for a remote headless browser service and uses established libraries like Playwright and Puppeteer.

headless-browser