The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill's primary interface is the browser-use CLI, which executes Python code provided via shell heredocs (e.g., browser-use <<'PY'). This architecture allows for the runtime generation and execution of arbitrary logic within the skill's environment.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to scrape and interact with arbitrary external websites, creating a significant ingestion surface for untrusted data.
Ingestion points: Data enters the agent's context through page_info(), DOM inspection via js(...), and accessibility tree extraction using cdp("Accessibility.getFullAXTree") in SKILL.md.
Boundary markers: The instructions do not specify the use of delimiters or specific prompts to ensure the agent disregards instructions that might be embedded in the HTML or AXTree of the pages it visits.
Capability inventory: The skill possesses capabilities including network access (via the browser), filesystem writes (via the recordings feature), and shell command execution (via the CLI wrapper itself).
Sanitization: No explicit sanitization or filtering logic is mentioned for the content retrieved from external sources before it is processed by the agent.
[COMMAND_EXECUTION]: The skill relies on executing the browser-use command-line utility to perform its tasks, including interacting with the browser daemon and managing remote browser instances.
[PRIVILEGE_ESCALATION]: The skill includes documentation for a mac-approve command and instructions for users to grant Accessibility permissions in macOS System Settings. While this requires manual user intervention, it facilitates granting the application high-level control over the operating system's user interface and other applications.
[EXTERNAL_DOWNLOADS]: The skill installs the browser-use package via uv as defined in its metadata. It also references external documentation and configuration guides hosted on GitHub and points users to a remote cloud browser service at cloud.browser-use.com for parallel or isolated tasks.

browser-use