The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill interacts with and reads data from a browser environment, which introduces a surface for indirect prompt injection if the automated browser accesses untrusted content. * Ingestion points: The skill reads application state and DOM content using the browser_eval and browser_snapshot tools, specifically checking document.body.innerText. * Boundary markers: There are no explicit delimiters or instructions provided to the agent to ignore or isolate instructions found within the processed browser data. * Capability inventory: The skill allows for JavaScript execution in the browser context via browser_eval, as well as simulating user interactions like clicking buttons and pasting text. * Sanitization: The skill does not describe any validation or sanitization of the data retrieved from the browser before it is returned to the agent context.
[COMMAND_EXECUTION]: The skill provides specific instructions for running local shell commands to manage the development environment. * Evidence: The documentation includes commands such as pnpm dev, nohup pnpm dev for background execution, and lsof to monitor network ports.
[DYNAMIC_EXECUTION]: The skill's primary functionality relies on executing dynamic JavaScript within the Electron browser target. * Evidence: The browser_eval pattern uses a self-invoking JavaScript function to query the DOM, manipulate the clipboard, and trigger UI events.

browser-automation