The Agent Skills Directory

[COMMAND_EXECUTION]: The eval command (including --stdin and -b variants) enables execution of arbitrary JavaScript within the browser's context (SKILL.md, references/commands.md).
[DATA_EXFILTRATION]: The state save command exports session tokens and cookies to a local file, which are stored in plaintext unless an encryption key is provided via AGENT_BROWSER_ENCRYPTION_KEY (SKILL.md, references/session-management.md).
[DATA_EXFILTRATION]: Commands like clipboard read and network requests provide mechanisms for extracting data from the host or active browser session (references/commands.md).
[PROMPT_INJECTION]: The skill is designed to ingest and process data from external, untrusted websites, creating a surface for indirect prompt injection.
Ingestion points: agent-browser snapshot, agent-browser get text, agent-browser network requests.
Boundary markers: The skill provides an opt-in AGENT_BROWSER_CONTENT_BOUNDARIES feature to wrap page content in nonces.
Capability inventory: eval, click, fill, network route, download, state save, clipboard write.
Sanitization: No automatic sanitization of web content is performed; the skill relies on the agent's internal guardrails and the optional boundary markers.
[EXTERNAL_DOWNLOADS]: The documentation provides instructions for installing the tool and its dependencies via standard package managers including npm, brew, and cargo (SKILL.md).

agent-browser