System prompts

What is System Prompting?

• System prompts customize the style and tone of Claude's responses
• They are plain strings passed to the API's create function
• They allow you to steer Claude's behavior in a particular direction

Key Use Case: Math Tutor Example

• Without system prompt: Claude gives direct, complete answers
• With system prompt: Claude guides students with hints and step-by-step guidance
• System prompts help create more interactive and educational experiences

How to Implement System Prompts

• Define a system prompt as a multi-line string
• Assign Claude a role in the first line (e. g. , "You are a patient math tutor")
• Pass the system prompt to the API using the system keyword argument in the create function call

Handling Optional System Prompts

• Add a system parameter to your chat function with a default value of None
• Create a params dictionary to hold API parameters
• Only add the system key to the params dictionary if a system prompt was actually provided
• Use **params syntax to unpack the dictionary when calling the create function
• This prevents errors from passing system=None to the API

Benefits

• Encourages desired behavior (e. g. , patient guidance instead of direct answers)
• Makes chat functions reusable across different scenarios
• Provides more control over response style without changing the user's input

Takeaways

• System prompts customize Claude's style and tone by assigning it a role and behavior guidelines
• System prompts enable interactive experiences like step-by-step tutoring instead of direct answers
• Pass system prompts to the API using the system keyword argument in the create function
• Handle optional system prompts by only adding the system key to the params dictionary if one is provided
• System prompts make chat functions reusable across different scenarios without modifying user input