The Agent Skills Directory

[DATA_EXFILTRATION]: The browser-cookies.js script allows the agent to retrieve and print all browser cookies for the active session, including sensitive authentication tokens and session identifiers, to the agent's console output.
[DATA_EXFILTRATION]: The browser-start.js script includes a --profile flag that copies the user's entire Chrome profile directory (containing history, saved passwords, and active sessions) to a local cache folder, exposing private user data to the agent's environment.
[DYNAMIC_EXECUTION]: The browser-eval.js tool enables the execution of arbitrary JavaScript code within the browser context via new AsyncFunction, which allows for programmatic manipulation of web pages but also provides a mechanism for executing untrusted scripts.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates an attack surface where malicious content on websites could influence agent behavior, as the agent ingests untrusted data while possessing powerful capabilities like cookie extraction and JS evaluation.
Ingestion points: browser-content.js and browser-hn-scraper.js fetch and parse external web content into the agent's context.
Boundary markers: The skill lacks explicit markers to distinguish between legitimate page content and potential instructions embedded in the HTML.
Capability inventory: The skill provides full browser control including script execution (browser-eval.js), secret extraction (browser-cookies.js), and navigation (browser-nav.js).
Sanitization: No sanitization or filtering is performed on the ingested web content before it is presented to the agent.

browser-tools