The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the belt CLI (and its variant infsh) along with standard shell utilities like jq and mkdir to automate browser tasks and manage output files. It also utilizes shell redirection to write extracted data to the local file system.
[REMOTE_CODE_EXECUTION]: The execute function allows the agent to run arbitrary JavaScript within the browser context. This capability is used to extract specific data and manipulate the DOM, which is a standard feature for web automation tools.
[DATA_EXFILTRATION]: The skill performs network operations to interact with web pages and includes functions to extract sensitive information such as page text, links, and session cookies for further processing.
[PROMPT_INJECTION]: The skill is designed to ingest and process data from external, untrusted websites, creating an attack surface for indirect prompt injection.
Ingestion points: Content from arbitrary URLs is brought into the agent's context through functions like open, snapshot, and execute (referenced in SKILL.md and references/commands.md).
Boundary markers: The instructions do not specify the use of delimiters or warnings to ignore instructions embedded in the retrieved web content.
Capability inventory: The agent has the ability to execute shell commands, run browser-side JavaScript, and write files to the local system.
Sanitization: No sanitization or validation of the fetched external content is described before it is returned to the agent.
[EXTERNAL_DOWNLOADS]: Fetches installation guidelines and documentation from the vendor's official GitHub repository (inference-sh/skills).

agent-browser