The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill's primary entry point is via npx agent-browser, which dynamically fetches and executes the automation engine from the npm registry at runtime.
[REMOTE_CODE_EXECUTION]: The eval command permits the execution of arbitrary JavaScript within the browser environment. The skill documentation explicitly supports and recommends Base64 encoding or stdin for these scripts to bypass shell escaping, which can obfuscate the code's intent.
[COMMAND_EXECUTION]: The skill provides extensive capabilities to run browser instances, including support for custom browser binaries via the --executable-path flag and the ability to load arbitrary browser extensions.
[DATA_EXFILTRATION]: When the --allow-file-access flag is used, the agent can navigate to file:// URLs, allowing it to read and extract sensitive local files via screenshots, PDF generation, or text extraction commands.
[PROMPT_INJECTION]: As the skill is designed to process external web content, it is susceptible to indirect prompt injection where malicious instructions on a webpage could influence the agent's behavior. The skill includes an optional --content-boundaries feature to help mitigate this risk.
[CREDENTIALS_UNSAFE]: The auth vault functionality stores sensitive credentials in a local state. While it supports encryption, the security of these secrets relies on the management of the AGENT_BROWSER_ENCRYPTION_KEY environment variable.

agent-browser