The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill provides an eval command (documented in SKILL.md and references/commands.md) that allows executing arbitrary JavaScript within the browser context. This is a powerful feature that could be misused to perform actions not explicitly defined by the agent's instructions.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the internet, creating a surface for injection attacks. \n
Ingestion points: agent-browser snapshot and agent-browser get text in SKILL.md.\n
Boundary markers: Provides an optional AGENT_BROWSER_CONTENT_BOUNDARIES environment variable to wrap output.\n
Capability inventory: Capabilities include file writing, interactive element manipulation, and script execution.\n
Sanitization: No specific sanitization of the extracted text content is mentioned in the instructions.\n- [DATA_EXFILTRATION]: The skill saves and loads authentication states, including cookies and localStorage, to local files (e.g., auth-state.json). Access to these files could lead to session hijacking if the host environment is not properly secured.\n- [EXTERNAL_DOWNLOADS]: The skill documentation recommends execution via npx, which results in the download and execution of the agent-browser package from a remote registry.\n- [COMMAND_EXECUTION]: The skill utilizes Bash to invoke its CLI commands, which includes parameters for complex configurations such as network proxies and custom browser executables.

agent-browser