Hello everyone. I confused with OpenAi agent, agent has to run chain of tools like (1)"read_table_properties" -> (2)"form record from user's message that is compatible with table properties" -> (*3)"write record". Major problem that agent do not correctly transforms data from user's message to table structure. Maybe that problem can be solved by creation OpenAi instance with prompt insade tool, but I am not sure are good solution. How do you think ?

Hello everyone. I've been working with OpenAI agent, the app I'm developing works with Notion integration and custom prompts, which clients can change on their own. In order to work with agent I've created BaseToolSpec. It works correct. However the final answer from llm generates for a long time and without consideration of the custom prompt. whislt not receiving large amount of data. There were also cases where llm doesn't even generate an answer on already tested cases whilst using the same tools.
The question is how can I make the agent work faster and fix the errors where the agent does not generate answers and does not consider system_prompt