corsairdev/corsair ? reverse-engineered prompt

I want a system that acts as a secure middle layer between my AI agent and different service integrations like Google Drive, Gmail, Slack, and GitHub. The key thing is I should be able to connect my agent to all these services without ever giving it my actual credentials. I want to control exactly what actions the agent is allowed to take, for example making sure that anything potentially destructive or sensitive—like sending emails—requires my explicit approval before it happens. It should also support multiple tenants so each client or team has their permissions and data separated. If I deny a request, nothing should happen, and the agent can only proceed if I approve. Include webhook support so integrations can notify my app through a single endpoint. Please make sure I can manage permission levels easily for each integration, letting some things run freely but requiring review for others. Look up current docs online if you need examples of integrations.