The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires the global installation of an unverified npm package agent-browser and downloads Chromium binaries via agent-browser install.
[COMMAND_EXECUTION]: Relies on the Bash tool to execute CLI commands for navigating websites and performing browser-based tasks.
[DATA_EXFILTRATION]: Provides specific instructions and commands to retrieve sensitive data from the browser, such as cookies get, storage local get, and authentication tokens extracted via JavaScript. This data could be exfiltrated using the available Bash capabilities.
[REMOTE_CODE_EXECUTION]: The skill features an eval command that allows arbitrary JavaScript execution within the browser context, which can be used to steal credentials or perform unauthorized actions on websites.
[PROMPT_INJECTION]: The skill has a significant exposure to indirect prompt injection as it processes data from untrusted external websites. * Ingestion points: Ingests web content via snapshot, get text, and get html as described in SKILL.md and references/ai-snapshot-workflow.md. * Boundary markers: There are no instructions to use delimiters or ignore embedded instructions when processing web data. * Capability inventory: The skill has access to shell commands (Bash) and arbitrary browser script execution (eval). * Sanitization: The documentation does not mention any sanitization or validation of the content retrieved from the web.

agent-browser