Granular structured way to prompt engineer

So rather than fine tuning, for our chatbot, we use langchain agents and tools. Each tool gets its own prompt as well as a global default prompt for the agent.

For things we need to adjust, specific and general tasks, we tend to add new instructions into each prompt for each tool. The prompts end up too long, the responses come slower and at one point, I think it will be harmful rather than useful.

Did anyone come up with a system or basic software to adjust and structure prompts? An example is, I want the LLM to make a search when there is a mention of the word “lately”, “currently” etc. and find the most recent information. Any way to systemize these kind of instructions?