The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill utilizes sudo within the sandbox environment to install required system packages for Chromium.
[REMOTE_CODE_EXECUTION]: The skill installs the agent-browser tool and its dependencies from npm during the sandbox environment setup.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted data from external URLs through browser automation, creating a surface for indirect prompt injection.
Ingestion points: The url parameter in screenshotUrl and snapshotUrl, and the data parameter in fillAndSubmitForm are used to navigate and interact with external websites.
Boundary markers: No boundary markers or warnings to ignore embedded instructions are present in the provided code templates.
Capability inventory: The skill has the capability to run arbitrary commands within the sandbox (runCommand), capture screenshots, and retrieve accessibility snapshots.
Sanitization: There is no evidence of content sanitization or filtering of the external data before it is returned to the agent context.

vercel-sandbox