Sometimes I have a hard time understanding how Agent is able to prototype its data schema and Google Login in a the first prompt and then, in a new chat, is unable to add a new column to a table.
As you can see from the screenshots, it appears to have a perfect understanding of the instructions. Something happens and it fails, I accept its suggestions, which seem coherent and since the application is not in production I am not worried about resetting the database.
However, it fails again and again, it announces that it will do the correct fix: add the “team_id” column to “user” but it doesn’t do it.
In the end I launched the SQL myself and everything worked again.
Along the way, I got 2 extra checkpoints…
Do you use any specific prompts to manipulate the database? Apparently Agent understands you well, but when you run it, it feels like it doesn’t know the current schema and the data it contains, it starts modifying without information, as if the DB were new, and it reaches the result with trial and error, which doesn’t always end well.
Sometimes I think it would be better to ask him to code SQL scripts and I’ll run them myself, but then you have to replicate it in the ORM.
I don’t know if you have to give him explicit instructions to “first verify the schema and data of the current DB.”
(screenshot instructions are in spanish)
