The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the osascript command to execute AppleScript, allowing it to control the host's Google Chrome application and run JavaScript in the active tab. This is listed as a fallback method for stubborn elements.
[DYNAMIC_EXECUTION]: The skill uses browser evaluate to execute dynamically generated JavaScript snippets within the browser context for form manipulation and interaction, such as clicking radio buttons or handling custom widgets.
[DATA_EXFILTRATION]: The skill reads sensitive information from USER.md and MEMORY.md (e.g., professional background, role, company details) and inputs this data into third-party survey websites, effectively transmitting local user data to external entities.
[INDIRECT_PROMPT_INJECTION]: By navigating to and interacting with external survey sites and emails, the skill is exposed to untrusted data that could contain malicious instructions. Ingestion points: Gmail messages and external survey pages (Arbolus, Guidepoint, GLG, etc.); Boundary markers: Mentions checking for AI-detection warnings; Capability inventory: File system read, browser automation, and shell execution; Sanitization: No explicit sanitization or escaping of external content is described before processing.
[CREDENTIALS_UNSAFE]: The skill leverages the user's persistent Chrome profile and Gmail access via MCP tools, which contain session cookies and authentication credentials for various services.

survey-passing