The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to install the agent-browser package from npm and provides source installation steps via a Git clone from Vercel Labs' official GitHub repository.
[COMMAND_EXECUTION]: Through the allowed-tools configuration, the skill grants access to the agent-browser CLI. This tool includes powerful capabilities such as:
JavaScript Evaluation: The eval and wait --fn commands allow the execution of arbitrary JavaScript within the browser's context.
File System Interaction: The skill can write files to the local system (via screenshot, pdf, state save, and trace stop) and read files for uploads (via upload and state load).
Network Manipulation: The network route commands enable request interception, blocking, and mocking.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) because it processes untrusted data from external websites.
Ingestion points: Data enters the agent context through commands like snapshot, get text, get html, and get attr (SKILL.md).
Boundary markers: There are no explicit instructions or delimiters provided to the agent to treat page content as untrusted or to ignore embedded instructions.
Capability inventory: The agent has the ability to browse any URL, write local files, and execute JavaScript in the browser (SKILL.md).
Sanitization: The skill does not implement sanitization or filtering of the content retrieved from the web before presenting it to the agent.

Agent Browser