AgentScript — IT Support Agent with Salesforce Agent Script

Built by Kapil Batra | SF Bolt YouTube Channel

This repository contains the complete source code for the IT Support Agent built using Salesforce Agent Script — demonstrated in the SF Bolt video series on Agentforce development.

What Is This?

This project demonstrates how to build a predictable, enterprise-grade Agentforce agent using Agent Script — Salesforce's pro-code scripting language for AI agents announced at Dreamforce 2025 (public beta November 2025).

Agent Script preserves the conversational intelligence of LLM-based agents and adds the predictability of programmatic execution. You define exactly which parts of your workflow execute deterministically and which parts leverage LLM reasoning. The result is hybrid reasoning — strict business logic where you need it, intelligent flexibility where you want it.

What the Agent Does

The IT Support Agent handles three support scenarios for Acme Corp employees:

Scenario	Behaviour
Password Reset	Verifies identity → checks MFA status → resets only after explicit YES confirmation
Software Access	Verifies identity → routes automatically for Engineering, requires manager approval for all other departments
Hardware Replacement	Verifies identity → checks warranty → discloses $250 cost if out of warranty → proceeds only after YES confirmation

All three scenarios enforce identity verification first — deterministically, in code, every single time.

What Agent Script Adds

According to the official Salesforce documentation, Agent Script lets you define:

Specific areas where the LLM is free to reason — intent classification, natural language understanding
Specific areas that execute deterministically — flows called in guaranteed sequence via -> (arrow)
Variables that reliably store state — persists across every conversation turn, not relying on LLM memory
Conditional expressions controlling execution paths — branching driven by data, not LLM judgment
Deterministic subagent transitions — routing in code, not as an LLM suggestion

The two core symbols:

->   Arrow = deterministic. Code runs exactly as written, every time.
|    Pipe  = prompt. This text goes to the LLM to reason about.

Repository Structure

AgentScript/
├── force-app/
│   └── main/
│       └── default/
│           ├── aiAuthoringBundles/
│           │   └── IT_Support_Agent/
│           │       └── IT_Support_Agent.agent        # Agent Script file
│           ├── flows/
│           │   ├── Verify_Identity.flow-meta.xml
│           │   ├── Check_MFA_Status.flow-meta.xml
│           │   ├── Reset_Password.flow-meta.xml
│           │   └── Raise_Support_Ticket.flow-meta.xml
│           └── objects/
│               ├── Contact/
│               │   └── fields/
│               │       ├── EmployeeNumber__c.field-meta.xml
│               │       ├── Has_MFA__c.field-meta.xml
│               │       ├── Under_Warranty__c.field-meta.xml
│               │       └── Password_Reset_Required__c.field-meta.xml
│               └── IT_Support_Request__c/
│                   ├── IT_Support_Request__c.object-meta.xml
│                   └── fields/
│                       ├── Contact__c.field-meta.xml
│                       ├── Issue_Type__c.field-meta.xml
│                       ├── Status__c.field-meta.xml
│                       ├── Notes__c.field-meta.xml
│                       ├── Requires_Manager_Approval__c.field-meta.xml
│                       └── Chargeble__c.field-meta.xml
├── scripts/
│   └── apex/
│       └── CreateTestContacts.apex
├── manifest/
│   └── package.xml
└── README.md

Prerequisites

Before deploying this project you need:

Salesforce Developer Edition org with Einstein and Agentforce enabled
Salesforce CLI v2 installed — verify with sf -v
VS Code with the Salesforce Extension Pack installed
Agentforce DX VS Code Extension installed
Prompt Template Manager permission set assigned to your user
An authorized SFDX project connected to your org

Custom Object: IT_Support_Request__c

Created to log all IT support requests raised by the agent.

Field Label	API Name	Type	Notes
Request Number	Name	Auto Number	Format: REQ-{0000}
Contact	Contact__c	Lookup (Contact)	Links ticket to employee
Issue Type	Issue_Type__c	Picklist	password_reset, software_access, hardware
Status	Status__c	Picklist	Open, Pending Approval, Escalated, Resolved
Notes	Notes__c	Long Text Area	Optional notes
Requires Manager Approval	Requires_Manager_Approval__c	Checkbox	Default: false
Chargeable	Chargeble__c	Checkbox	Default: false

Note: Chargeble__c has a typo in the field label — this matches what was deployed to the org. Do not rename without updating the Raise_Support_Ticket flow accordingly.

Custom Fields on Contact

Field Label	API Name	Type	Default	Purpose
Employee Number	EmployeeNumber__c	Text (20)	—	Unique employee identifier used for identity verification
Has MFA	Has_MFA__c	Checkbox	false	Whether employee has MFA enabled — gates password reset
Under Warranty	Under_Warranty__c	Checkbox	true	Whether employee device is under warranty
Password Reset Required	Password_Reset_Required__c	Checkbox	false	Updated to false when Reset_Password flow runs

Flows

1. Verify_Identity

Purpose: Looks up a Contact record by EmployeeNumber__c and returns identity confirmation, Contact Id, and department.

Variable	Direction	Type	Description
employee_id_input	Input	Text	Employee ID provided by the user
verified	Output	Boolean	True if a matching Contact was found
contact_id	Output	Text	Salesforce Contact Id
department	Output	Text	Contact.Department field value

Logic: Get Records on Contact where EmployeeNumber__c = employee_id_input. On success — set verified = true, map Id and Department. On fault — set verified = false.

2. Check_MFA_Status

Purpose: Checks the Has_MFA__c field on the verified Contact record.

Variable	Direction	Type	Description
contact_id	Input	Text	Salesforce Contact Id
has_mfa	Output	Boolean	Value of Has_MFA__c on the Contact

Logic: Get Records on Contact where Id = contact_id. Map Has_MFA__c to output. On fault — set has_mfa = false.

3. Reset_Password

Purpose: Simulates a password reset by updating Password_Reset_Required__c to false on the Contact record.

Variable	Direction	Type	Description
contact_id	Input	Text	Salesforce Contact Id
reset_successful	Output	Boolean	True if record update succeeded

Logic: Update Contact record — set Password_Reset_Required__c = false. On success — set reset_successful = true. On fault — set reset_successful = false.

4. Raise_Support_Ticket

Purpose: Creates an IT_Support_Request__c record with routing logic based on issue type and department.

Variable	Direction	Type	Description
contact_id	Input	Text	Salesforce Contact Id
issue_type	Input	Text	One of: password_reset, software_access, hardware
department	Input	Text	Employee department for routing
ticket_number	Output	Text	Created record Id
requires_approval	Output	Boolean	True if manager approval required
chargeable	Output	Boolean	True if hardware replacement incurs cost

Routing logic:

Issue Type	Condition	Status	Requires Approval	Chargeable
password_reset	Any	Escalated	false	false
software_access	Department = Engineering	Open	false	false
software_access	Department ≠ Engineering	Pending Approval	true	false
hardware	Under_Warranty__c = true	Open	false	false
hardware	Under_Warranty__c = false	Open	false	true

Agent Script — IT_Support_Agent.agent

Block Structure (in order)

Block	Purpose
system	Global agent personality, welcome and error messages
config	Agent metadata — API name, type, role, company context
variables	Six global variables persisting across all conversation turns
language	Supported locales
start_agent	Entry point — routes to correct subagent based on verified state
subagent	Core IT support logic — actions, reasoning instructions, tools

Variables

Name	Type	Default	Description
contact_id	mutable string	""	Salesforce Contact Id of the verified employee
verified	mutable boolean	False	Whether identity has been confirmed this session
has_mfa	mutable boolean	False	Whether the employee has MFA enabled
department	mutable string	""	Employee department for routing decisions
requested_action	mutable string	""	What the employee explicitly asked for
action_confirmed	mutable boolean	False	Whether employee confirmed an irreversible action

Reasoning Flow

Every message → start_agent
   → if not verified → prompt for employee ID
   → if verified     → transition to IT_Support_Handler

IT_Support_Handler
   → DETERMINISTIC: run Verify_Identity flow
   → LLM: classify intent → set requested_action
   → DETERMINISTIC: branch on requested_action value
      → "question"        → LLM answers, no action taken
      → "password_reset"  → confirmation gate → check MFA → reset
      → "software_access" → confirmation gate → route by department
      → "hardware"        → warranty check → cost disclosure → confirm

How to Deploy

Step 1 — Clone the repository

git clone https://github.com/batra-kapil/AgentScript.git
cd AgentScript

Step 2 — Authorize your org

sf org login web --alias your-org-alias

Step 3 — Deploy custom fields first

sf project deploy start \
  --source-dir force-app/main/default/objects \
  --target-org your-org-alias

Step 4 — Deploy flows

sf project deploy start \
  --source-dir force-app/main/default/flows \
  --target-org your-org-alias

All four flows must deploy and activate successfully before the agent can reference them.

Step 5 — Deploy the Agent Script

sf project deploy start \
  --source-dir force-app/main/default/aiAuthoringBundles \
  --target-org your-org-alias

Step 6 — Create test Contact records

Run in Developer Console → Execute Anonymous:

List<Contact> contacts = new List<Contact>{
    new Contact(
        FirstName = 'Arjun',
        LastName = 'Kapoor',
        Department = 'Engineering',
        EmployeeNumber__c = 'EMP001',
        Has_MFA__c = true,
        Under_Warranty__c = true,
        Password_Reset_Required__c = true
    ),
    new Contact(
        FirstName = 'Sneha',
        LastName = 'Iyer',
        Department = 'Finance',
        EmployeeNumber__c = 'EMP002',
        Has_MFA__c = false,
        Under_Warranty__c = true,
        Password_Reset_Required__c = false
    ),
    new Contact(
        FirstName = 'Vikram',
        LastName = 'Nair',
        Department = 'HR',
        EmployeeNumber__c = 'EMP003',
        Has_MFA__c = true,
        Under_Warranty__c = false,
        Password_Reset_Required__c = false
    )
};
insert contacts;
System.debug('Test contacts created successfully');

Step 7 — Activate the agent

Go to Setup → Agentforce Agents → IT Support Agent → Activate

Step 8 — Test in Agentforce Builder

Open Agentforce Builder → click Preview and run these conversations:

Test 1 — Password reset happy path (EMP001 — has MFA)

My employee ID is EMP001. I need a password reset.
→ YES

Expected: Identity verified → MFA confirmed → password reset after YES.

Test 2 — Software access, Finance department (EMP002)

My employee ID is EMP002. I need access to the Finance reporting tool.
→ YES

Expected: Manager approval required — Sneha is in Finance.

Test 3 — Hardware, out of warranty (EMP003)

My employee ID is EMP003. My laptop is broken.
→ YES

Expected: $250 cost disclosed before confirmation — Vikram's device is out of warranty.

Test 4 — Question only, no action taken

My employee ID is EMP001. How long does a password reset usually take?

Expected: Agent answers the question. No reset fires.

Debugging

Interaction Summary — Agentforce Builder

Open Agentforce Builder → run a test conversation → check the Interaction Summary panel. Shows each step in sequence: Input, Reasoning, Topic Transitions, Actions, Output Evaluation.

Trace Tab — Agentforce Builder

Expand the Trace tab at the bottom of Agentforce Builder for millisecond-level timing and execution detail.

Agent Tracer — VS Code

Open the Agent Tracer tab in the Agentforce DX VS Code extension. Shows full execution plan: User Input, Topic Selected, Variable Updates, Tools Enabled, Reasoning steps, Topic Transitions. Click any conversation turn to view the complete raw JSON execution trace.

Event Logs — Production Monitoring

Go to Setup → Agentforce Agents → IT Support Agent → Edit → Enable "Enrich event logs with conversation data"

Then query via SOQL:

SELECT ConversationDefinitionId, EventDateTime, EventLabel, 
       EventTarget, EventTargetType, StepType
FROM ConversationDefinitionEventLog
WHERE CreatedDate = TODAY

Known Issues

Chargeble__c has a typo in the field label — this is intentional to match the deployed org field name. Do not rename without updating the Raise_Support_Ticket flow.
EmployeeNumber__c is a custom field — the standard Contact object EmployeeNumber field was not available in this org configuration.
Deploy order matters — objects before flows before agent. Deploying out of order will cause atomic rollback failures.

Resources

About SF Bolt

SF Bolt is a YouTube channel and LinkedIn brand focused on Salesforce and AI content — built by Kapil Batra, Salesforce MVP and Certified Application Architect.

YouTube: SF Bolt
LinkedIn: Kapil Batra
GitHub: batra-kapil

If this helped you, star the repo and subscribe to SF Bolt for more Agentforce content every week.

Name		Name	Last commit message	Last commit date
Latest commit History 2 Commits
.claude		.claude
.husky		.husky
.vscode		.vscode
config		config
force-app/main/default		force-app/main/default
scripts		scripts
.forceignore		.forceignore
.gitignore		.gitignore
.prettierignore		.prettierignore
.prettierrc		.prettierrc
README.md		README.md
eslint.config.js		eslint.config.js
jest.config.js		jest.config.js
package.json		package.json
sfdx-project.json		sfdx-project.json

Folders and files

Latest commit

History

Repository files navigation

AgentScript — IT Support Agent with Salesforce Agent Script

What Is This?

What the Agent Does

What Agent Script Adds

Repository Structure

Prerequisites

Custom Object: IT_Support_Request__c

Custom Fields on Contact

Flows

1. Verify_Identity

2. Check_MFA_Status

3. Reset_Password

4. Raise_Support_Ticket

Agent Script — IT_Support_Agent.agent

Block Structure (in order)

Variables

Reasoning Flow

How to Deploy

Step 1 — Clone the repository

Step 2 — Authorize your org

Step 3 — Deploy custom fields first

Step 4 — Deploy flows

Step 5 — Deploy the Agent Script

Step 6 — Create test Contact records

Step 7 — Activate the agent

Step 8 — Test in Agentforce Builder

Debugging

Interaction Summary — Agentforce Builder

Trace Tab — Agentforce Builder

Agent Tracer — VS Code

Event Logs — Production Monitoring

Known Issues

Resources

About SF Bolt

About

Resources

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages