A system prompt sets an LLM application’s role, response requirements, and behavioral boundaries. In a documentation assistant, it tells the model to answer from retrieved documents, state when information is missing, and use a consistent format. These instructions help standardize responses, but they are not a substitute for input validation or application-level error handling. section 13.1.1 section 12.3
Specify the rules that matter to the application:
- Role and scope: identify the intended audience and limit answers to documented features. section 13.1.1
- Uncertainty and safety: prohibit speculation and disclosure of credentials or internal instructions; say when the documentation lacks an answer. section 13.1.1 section 12.3
- Output: require a direct answer, relevant steps or code, and source citations in a predictable format. section 13.1.1
- Runtime context: provide retrieved documentation, the user’s question, and relevant session or environment information. section 13.1.1
From a vague instruction to an operational prompt
“You are a helpful assistant that answers questions about company documentation” leaves scope, uncertainty, and output format unspecified. A more useful documentation-assistant prompt defines a technical audience, restricts answers to provided context, redirects out-of-scope questions, and requires document-and-section citations. section 13.1.1
In a retrieval-augmented application, format retrieved documents consistently before inserting them into the prompt. Numbered documents and source metadata make the context easier to interpret and support citations in the response. Return sources and errors as structured results so the interface can display them separately from answer text. section 12.3
Listing 13.1 Creating managed prompts in Langfuse with versioning
langfuse.create_prompt(
name="dakkabot-system", #1
type="chat", #2
prompt=[
{
"role": "system",
"content": """You are DakkaBot, DataKrypt's expert
developer assistant. Based on the provided DataKrypt documentation,
answer the user's question accurately and helpfully.
**Documentation Context:**
{{context}}
**User Question:** {{query}}
**Instructions:**
- Provide accurate, detailed answers based on the documentation
- Include specific steps, code examples, or configurations
when relevant
- If the question requires multiple services or
procedures, explain the complete workflow
- Use proper formatting for code blocks and lists
- If you're not sure about something, say so rather than guessing
- Stay focused on DataKrypt-specific information
**Answer:**"""
}
],
labels=["production"], #3
config={
"model": "gemini-2.5-flash",
"temperature": 0.0,
"max_tokens": 1000
}, #4
tags=["dakkabot", "rag", "documentation"] #5
)
The managed chat-prompt listing shows a system-role message with {{context}} and {{query}} placeholders, keeping task instructions together while supplying retrieval results and the question at runtime. section 13.1.2
Treat production prompts as versioned application logic rather than hardcoded strings. Langfuse can store a prompt by name, label versions for environments, and attach model configuration. The application fetches the selected version, compiles its runtime variables, and records the version with the response. This permits prompt updates without redeploying application code and makes changes traceable. section 13.1 section 13.1.2
