Much safer as the technique ensures the AI agent never even sees the creds vs the CLI the inference/context can see them which opens you up to a whole class of security risk.

https://support.1password.com/1password-claude-security/