Many places isnt always good, it might confuse LLM, try structuring your prompt with markdown, optimize so it has neccessary instructions. something like:

# IDENTITY AND PURPOSE
Yada yada

# REQUIRED BEHAVIOR
1. **Step 1 (give step actual name)**: yada yada
2. **Step 2 (give step actual name)**: yada yada
...
# OUTPUT FORMAT (EVERY MESSAGE)
If sing structured output, else something similiar
# ADDITIONAL CONSTRAINTS
Tell model what to not do and similiar