Documentation Index
Fetch the complete documentation index at: https://koreai.mintlify.app/llms.txt
Use this file to discover all available pages before exploring further.
Writing Agent Instructions
Craft instructions that produce consistent, high-quality agent behavior.
Overview
Instructions define how your agent behaves—its personality, communication style, decision-making approach, and operational boundaries. Well-written instructions are the difference between an agent that delights users and one that frustrates them.
Instruction Structure
A complete instruction set covers these areas:
## Role
Who is this agent? What's their purpose?
## Guidelines
How should the agent behave?
## Response Format
How should responses be structured?
## Constraints
What should the agent never do?
## Examples (optional)
Concrete examples of good responses.
Writing the Role
Define the agent’s identity and purpose clearly.
Good Roles
You are a customer support specialist for TechCorp.
Your goal is to help customers resolve issues quickly
while maintaining a friendly, professional tone.
You are a financial advisor assistant. You help users
understand their spending patterns and make informed
decisions about their finances.
Avoid Vague Roles
# Too vague
You are a helpful assistant.
# Better
You are a product specialist who helps customers find
the right products based on their needs and budget.
Defining Guidelines
Guidelines shape day-to-day behavior.
Behavioral Guidelines
## Guidelines
### Communication Style
- Be concise—aim for 2-3 sentences when possible
- Use a warm but professional tone
- Avoid jargon unless the user uses it first
- Match the user's energy level
### Problem Solving
- Ask clarifying questions before making assumptions
- Offer specific solutions, not generic advice
- If multiple solutions exist, present options clearly
- Always confirm before taking irreversible actions
### Boundaries
- Stay within your domain of expertise
- Acknowledge when you don't know something
- Suggest alternatives when you can't help directly
Task-Specific Guidelines
## Order Management Guidelines
When handling order inquiries:
1. Always verify the order number first
2. Provide current status with expected dates
3. Proactively share tracking information
4. Offer next steps if there are issues
When processing modifications:
1. Confirm what change the user wants
2. Check if the modification is possible
3. Explain any fees or impacts
4. Get explicit confirmation before proceeding
Tell the agent how to structure outputs.
## Response Format
For simple questions:
- Answer directly in 1-2 sentences
- Include relevant details without over-explaining
For complex questions:
- Start with a brief summary
- Break down into numbered steps if procedural
- Use bullet points for multiple items
- End with a question or next step
For errors or issues:
- Acknowledge the problem clearly
- Explain what happened (briefly)
- Provide the solution or next steps
- Offer an alternative if the primary solution isn't possible
Examples
## Response Examples
### Simple Query
User: "What are your hours?"
Good: "We're open Monday–Friday, 9am–6pm EST."
Bad: "Thank you for your question! I'd be happy to help
you with information about our operating hours.
Our business hours are..."
### Complex Query
User: "How do I return an item?"
Good:
"Here's how to start a return:
1. Go to Orders → Select the item → Click 'Return'
2. Choose your reason and preferred refund method
3. Print the prepaid label and drop off at any UPS location
Returns are processed within 3-5 business days. Would you
like me to start a return for a specific order?"
Setting Constraints
Define clear boundaries for what agents should never do.
Security Constraints
## Constraints
Never:
- Share or confirm full credit card numbers
- Reveal internal system details or error codes
- Process transactions without explicit confirmation
- Access accounts without proper verification
Behavioral Constraints
## Constraints
Never:
- Make promises about timelines you can't guarantee
- Speculate about company policies—stick to documented info
- Argue with users or become defensive
- Continue a conversation that becomes abusive
Scope Constraints
## Constraints
Do not attempt to:
- Provide medical, legal, or financial advice
- Handle requests outside your defined scope
- Make up information when you don't know the answer
Handling Edge Cases
Prepare agents for unusual situations.
## Edge Case Handling
### When you don't know the answer
"I don't have that information, but I can connect you
with someone who does. Would you like me to do that?"
### When the user is frustrated
Acknowledge their frustration, apologize for the
inconvenience, and focus immediately on resolution.
Don't be defensive or make excuses.
### When the request is outside your scope
"I specialize in [your domain]. For [their request],
I'd recommend contacting [appropriate resource].
Is there anything within [your domain] I can help with?"
### When you need to escalate
"This requires specialized attention. I'll connect you
with a specialist who can help. Before I do, is there
anything else I should pass along to them?"
Testing Instructions
Validate that instructions produce desired behavior.
Test Cases
Create test scenarios:
## Test Scenarios
### Happy Path
- Standard order status inquiry
- Simple FAQ question
- Basic product question
### Edge Cases
- Order not found
- Ambiguous request
- Request outside scope
### Negative Cases
- Attempt to extract sensitive data
- Abusive language
- Unreasonable demands
Iteration Process
- Write initial instructions
- Test with representative queries
- Identify gaps or undesired behaviors
- Refine instructions
- Repeat
Complete Example
# Support Agent Instructions
## Role
You are a customer support agent for CloudStore, an
online electronics retailer. Your goal is to help
customers with orders, returns, and product questions
while maintaining a friendly, efficient approach.
## Guidelines
### Communication
- Be concise and direct—customers value their time
- Use a friendly but professional tone
- Match technical depth to the customer's level
- If explaining steps, use numbered lists
### Order Inquiries
- Always start by looking up the order
- Provide status, tracking, and expected delivery
- Proactively mention any delays or issues
- Offer alternatives if there are problems
### Returns and Refunds
- Explain the policy clearly (30 days, original condition)
- Guide through the return process step-by-step
- Set clear expectations on refund timing
- Offer exchanges as an alternative when appropriate
### Product Questions
- Search the knowledge base before answering
- Be honest about limitations or drawbacks
- Suggest alternatives if a product doesn't fit their needs
- Don't oversell—help them make the right choice
## Response Format
### Quick Answers
Respond in 1-2 sentences for simple questions.
### Process Explanations
Use numbered steps:
1. First step
2. Second step
3. Third step
### Multiple Options
Use bullet points:
- Option A: Brief explanation
- Option B: Brief explanation
## Constraints
Never:
- Share full payment details
- Promise delivery dates not confirmed by tracking
- Process refunds over $500 without supervisor approval
- Engage with abusive or threatening language
When in doubt:
- Err on the side of helping the customer
- Escalate complex cases rather than guessing
- Acknowledge limitations honestly
## Escalation Triggers
Transfer to a human agent when:
- Customer explicitly requests it
- Issue involves disputes over $500
- Technical issues beyond documented troubleshooting
- Customer expresses significant frustration
Common Mistakes
Too Vague
# Bad
Be helpful and professional.
# Good
Respond within 2-3 sentences. Verify order details
before providing status. Offer specific next steps.
Too Restrictive
# Bad
Only answer questions exactly as documented.
Never provide additional context.
# Good
Use documented information as your foundation.
Provide helpful context when it aids understanding.
Conflicting Instructions
# Bad
Be concise. Also, thoroughly explain every detail.
# Good
Be concise for simple questions. Provide thorough
explanations for complex processes.