ConditionalLogicPatterns

Overview

This recipe demonstrates how to write if/else logic in Agent Script. The agent evaluates a loan application with an if / else if / else chain and combines checks with and, or, and not.

Agent Flow

%%{init: {'theme':'neutral'}}%%
graph TD
    A[Start] --> B[start_agent agent_router]
    B --> C[Transition to loan_evaluation]
    C --> D[Set max_loan]
    D --> E{Any details missing?}
    E -->|Yes| F[Ask for missing details]
    F --> G[LLM calls collect_application]
    G --> D
    E -->|No| H{credit_score < 580?}
    H -->|Yes| I[Rejected]
    H -->|No| J{loan_amount > max_loan?}
    J -->|Yes| I
    J -->|No| K{credit_score < 670?}
    K -->|Yes| L{Collateral and 2+ years?}
    L -->|Yes| M[Approved at 8.5%]
    L -->|No| N[Review needed]
    K -->|No| O{credit_score < 740?}
    O -->|Yes| P[Approved at 6.5%]
    O -->|No| Q[Approved at 4.5%]

Key Concepts

• if / else if / else: Check conditions in order and run only the first branch that matches
• Combined conditions: Join checks with and, or, and not
• Order of branches: Each branch only sees the cases that earlier branches ruled out
• Calculated values: Store the result of an expression in a variable and reuse it in conditions

How It Works

Calculating a Limit

set @variables.max_loan = @variables.annual_income + @variables.annual_income + @variables.annual_income

Arithmetic supports + and -. To compare the loan with three times the income, add the income three times and store the result.

Checking for Missing Details with or

if @variables.applicant_name == "" or @variables.credit_score == 0 or @variables.annual_income == 0 or @variables.loan_amount == 0:
   set @variables.approval_status = ""
   | Ask for any missing details: name, credit score, annual income, loan amount, years at current employer, and whether they have collateral. Save them with {!@actions.collect_application}.

The first branch matches if any required detail is missing, and asks the user for it.

Evaluating the Application with else if

else if @variables.credit_score < 580:
   set @variables.approval_status = "rejected"
   set @variables.interest_rate = 0
else if @variables.loan_amount > @variables.max_loan:
   set @variables.approval_status = "rejected"
   set @variables.interest_rate = 0
else if @variables.credit_score < 670 and @variables.has_collateral and @variables.employment_years >= 2:
   set @variables.approval_status = "approved"
   set @variables.interest_rate = 8.5
else if @variables.credit_score < 670:
   set @variables.approval_status = "review_needed"
   set @variables.interest_rate = 0

Only the first matching branch runs. Earlier branches already handle scores below 580 and loans above max_loan, so credit_score < 670 needs no lower bound. Fair credit has two branches. The first uses and to approve applicants with collateral and two years at their job. The second sends everyone else to review.

Ending the Chain with else

else if @variables.credit_score < 740:
   set @variables.approval_status = "approved"
   set @variables.interest_rate = 6.5
else:
   set @variables.approval_status = "approved"
   set @variables.interest_rate = 4.5

else covers everything the earlier branches didn't match, which here is a credit score of 740 or more.

Key Code Snippets

Complete Decision Chain

if @variables.applicant_name == "" or @variables.credit_score == 0 or @variables.annual_income == 0 or @variables.loan_amount == 0:
   set @variables.approval_status = ""
   | Ask for any missing details: name, credit score, annual income, loan amount, years at current employer, and whether they have collateral. Save them with {!@actions.collect_application}.
else if @variables.credit_score < 580:
   set @variables.approval_status = "rejected"
   set @variables.interest_rate = 0
else if @variables.loan_amount > @variables.max_loan:
   set @variables.approval_status = "rejected"
   set @variables.interest_rate = 0
else if @variables.credit_score < 670 and @variables.has_collateral and @variables.employment_years >= 2:
   set @variables.approval_status = "approved"
   set @variables.interest_rate = 8.5
else if @variables.credit_score < 670:
   set @variables.approval_status = "review_needed"
   set @variables.interest_rate = 0
else if @variables.credit_score < 740:
   set @variables.approval_status = "approved"
   set @variables.interest_rate = 6.5
else:
   set @variables.approval_status = "approved"
   set @variables.interest_rate = 4.5

Collecting the Application

collect_application: @utils.setVariables
   description: "Save the applicant's loan application details from the conversation"
   with applicant_name = ...
   with credit_score = ...
   with annual_income = ...
   with loan_amount = ...
   with employment_years = ...
   with has_collateral = ...

Try It Out

Example Interaction

Agent: Welcome! I'll help evaluate your loan application.

User: I'd like a loan of 50000.

Agent: To evaluate your loan application, I need a few more details:
       - Your full name
       - Your credit score
       - Your annual income
       - How many years you have been with your current employer
       - Whether you have collateral

User: I'm Ben. Credit score 620, annual income 50000, 1 year at my job, no collateral.

Agent: Thank you, Ben. Your loan application is now under review. A loan officer will review your information and contact you with the next steps.

Behind the Scenes:

1. The first message has no name or credit score, so the first branch asks for the missing details
2. The LLM saves the details with collect_application
3. The instructions run again: max_loan is 150000 and the first branch no longer matches
4. A credit score of 620 is under 670, and without collateral the review branch matches

What's Next

• TemplateExpressions: Use conditions inside {!...} expressions
• AvailableWhenFiltering: Use conditions to control which actions are available
• ErrorHandling: Validate inputs with guard clauses

Testing

Test Case 1: Excellent Credit

• Input: "I'm Ana. Credit score 760, annual income 90000, I want a 120000 loan, 4 years at my job, no collateral."
• Expected: Approved at 4.5%

Test Case 2: Fair Credit Without Collateral

• Input: "I'm Ben. Credit score 620, annual income 50000, I want a 100000 loan, 1 year at my job, no collateral."
• Expected: Review needed

Test Case 3: Low Credit Score

• Input: "I'm Cara. Credit score 550, annual income 60000, I want a 20000 loan, 6 years at my job, I have collateral."
• Expected: Rejected

Test Case 4: Loan Over Three Times Income

• Input: "I'm Dan. Credit score 700, annual income 30000, I want a 120000 loan, 5 years at my job, I have collateral."
• Expected: Rejected

Test Case 5: Missing Details

• Input: "I'd like a loan of 50000."
• Expected: Agent asks for the missing details

Notes

• The loan rules and interest rates are illustrative only
• An if can't contain another if. Use an else if chain or combine the checks with and and or
• Arithmetic supports + and - only
• Conditions work in if statements, {!...} template expressions, and available when filters