The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides an eval command that allows the execution of arbitrary JavaScript within the browser context. This includes support for Base64-encoded payloads (-b flag) and input via stdin, which can be used to execute unverified logic or bypass shell escaping.
[DATA_EXFILTRATION]: The skill includes capabilities to access local host files using the file:// protocol when the --allow-file-access flag is used. Additionally, the state save command allows exporting sensitive session information, including cookies and local storage, to JSON files on the local filesystem.
[PROMPT_INJECTION]: The skill is highly susceptible to indirect prompt injection. The agent navigates to arbitrary URLs and processes the resulting page content via the snapshot command. A malicious website could embed instructions in the DOM that the agent might follow, leveraging the skill's powerful automation features (like fill with environment variables or eval) for unauthorized actions. Ingestion points: agent-browser snapshot (DOM content). Boundary markers: None provided in instructions. Capability inventory: Subprocess calls, file-write (state save, pdf, screenshot), network operations, and eval. Sanitization: None detected.
[COMMAND_EXECUTION]: The skill provides extensive control over browser sessions through the agent-browser CLI, including network request interception, device emulation, and automated UI interaction.

agent-browser