Sign your agent in: authorize LLM tools with the code you already have
How we let our Rails chatbot reuse the authorization we already had, by treating the agent as just another user and the LLM as just another response format. About 750 lines, no shadow API, no second auth system.
