The Agent Skills Directory

[SAFE]: The documentation includes a dedicated section on prompt injection mitigation in resources/per-turn-context-and-cache-management.md. It advises developers to sanitize user-controlled inputs by collapsing whitespace, neutralizing XML closing tags, and capping input length to prevent sandbox escape.
[SAFE]: The skill enforces a security boundary between the client and server. In resources/extending-tool-registry.md, it mandates that tools requiring credentials, secrets, or privileged database access must execute only on the server, adhering to the principle of least privilege.
[COMMAND_EXECUTION]: The verification section in resources/extending-tool-registry.md mentions testing the agent using npx -y agent-browser. This is a standard developer workflow for testing browser-based agents and does not represent a malicious execution pattern in this context.
[EXTERNAL_DOWNLOADS]: The skill references official documentation and guidelines from Arize and Anthropic (e.g., Claude prompt engineering best practices) to ensure secure and efficient agent behavior.

phoenix-pxi