The Agent Skills Directory

[COMMAND_EXECUTION]: The start command in browser.mjs launches Google Chrome or Chromium via node:child_process. It also utilizes system utilities like lsof, taskkill, pgrep, and tasklist for process management and diagnostics.
[REMOTE_CODE_EXECUTION]: The evaluate command in browser.mjs enables the execution of arbitrary JavaScript within the browser's context using page.evaluate(). This is a core feature for web automation and requires the explicit use of the --unsafe-allow evaluate flag.
[DATA_EXFILTRATION]: The network and console commands in browser.mjs allow the capture and retrieval of page logs and network request metadata, which may contain sensitive information present on visited websites.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) because it processes untrusted content from web pages.
Ingestion points: Untrusted data enters the agent context via snapshot, get-dom, and text commands in browser.mjs.
Boundary markers: The instructions do not define specific delimiters for untrusted web content.
Capability inventory: The skill provides significant capabilities including click, type, evaluate, and navigate in browser.mjs, and skills install in skill-install.mjs.
Sanitization: There is no explicit sanitization or filtering of web content before it is returned to the agent.

browser