The Agent Skills Directory

[DATA_EXPOSURE]: The skill provides instructions for the agent to access and copy sensitive browser profile data, including Local State and Cookies files from the user's primary Google Chrome directory (~/Library/Application Support/Google/Chrome). This data is used to "seed" automation profiles with active session information.
[COMMAND_EXECUTION]: The skill utilizes AppleScript to drive Google Chrome, which allows for the execution of arbitrary JavaScript within the context of specific browser tabs. It also uses Bash for secret management, specifically piping data to an external vault via vault set --stdin.
[DYNAMIC_EXECUTION]: Browser interactions are performed using page.evaluate via CDP and AppleScript's execute javascript command, which involve running dynamically generated or file-based code within the target web pages.
[INDIRECT_PROMPT_INJECTION]: The skill interacts with and extracts data from external third-party vendor dashboards (such as Cloudflare, Azure, and GitHub), creating an attack surface for indirect prompt injection.
Ingestion points: External DOM content and HTML metadata extracted during browser automation steps.
Boundary markers: Employs assertIdentity and waitForLogin functions to verify the account context before performing mutations.
Capability inventory: Utilizes Bash, Read, Grep, and Glob tools, alongside AppleScript and CDP-based browser control.
Sanitization: Instructs the agent to pipe secrets directly to a vault via standard input to prevent exposure in logs or conversation transcripts.

dashboard-forge