The Agent Skills Directory

[COMMAND_EXECUTION]: The provided code patterns use sandbox.runCommand to execute shell commands within ephemeral microVMs, including the use of sudo dnf for installing system dependencies and npm/npx for setting up automation tools.
[EXTERNAL_DOWNLOADS]: Facilitates the installation of the @vercel/sandbox package and the agent-browser tool from official package registries.
[PROMPT_INJECTION]: The skill implements a pattern for processing external web content (via URLs) which constitutes an indirect prompt injection surface. Evidence: 1. Ingestion points: The url parameter in functions such as screenshotUrl and snapshotUrl in SKILL.md. 2. Boundary markers: No explicit delimiters or instructions are provided to the agent to distinguish between its instructions and the data being processed. 3. Capability inventory: The sandbox environment allows for shell command execution, file system access, and complex browser interactions. 4. Sanitization: No sanitization of the external page content is performed before processing.

vercel-sandbox