The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is entirely based on executing the agent-browser command-line interface to perform system-level browser automation, including navigating to URLs, taking screenshots, and interacting with page elements.
[DATA_EXFILTRATION]: The tool includes commands to access sensitive browser session data, such as agent-browser cookies and agent-browser storage local. While these are functional requirements for testing and automation, they could be leveraged to extract authentication tokens or session information if used maliciously.
[REMOTE_CODE_EXECUTION]: The agent-browser eval and agent-browser wait --fn commands allow for the execution of arbitrary JavaScript code within the browser context. This provides a mechanism for dynamic code execution on any website the agent visits.
[PROMPT_INJECTION]: As a tool that ingests and processes content from external websites via agent-browser snapshot and get commands, it is susceptible to indirect prompt injection.
Ingestion points: Untrusted data enters the agent context through page snapshots and text extraction in SKILL.md.
Boundary markers: There are no explicit instructions or delimiters mentioned to prevent the agent from following instructions embedded within the processed web content.
Capability inventory: The tool possesses high-privilege capabilities including form filling (fill), file uploads (upload), JavaScript execution (eval), and session data access (cookies).
Sanitization: The skill documentation does not outline any sanitization or filtering of the extracted web content before it is processed by the agent.

agent-browser