Skip to main content
A Playbook is a procedure document that AI Agent uses to handle employee inquiries. Based on registered Playbooks, AI Agent guides requesters through the appropriate procedure. There are two ways to register Playbooks: file upload and inline creation.

​
Register via file upload

Use file upload to register multiple Playbooks at once.
Supported file formats are Markdown (.md), plain text (.txt), CSV (.csv), Word (.docx), and PDF (.pdf). Each file can be up to 5MB, and you can upload up to 50 files per operation. Each file must start with YAML frontmatter specifying procedure_id, name, description, category, and subcategory.
1

Open the Playbook management screen

From the AI Helpdesk admin screen, select AI Playbook.
2

Select files

Click Upload Procedures and select the files you want to register. You can select multiple files at once.
3

Review and confirm

After upload, each file’s frontmatter and body are parsed, and any errors are shown per file. Review the content and click Upload.
If multiple files share the same procedure_id, only the first one loaded is registered. Review your files before uploading.

​
Register via inline creation

Create a Playbook directly in the WebUI text editor.
1

Open the Playbook management screen

From the AI Helpdesk admin screen, select AI Playbook.
2

Start creating

Click Add Procedure to open the creation form.
3

Write the Playbook

Enter the Playbook content in the text editor. For step notation, see Playbook step notation below.Screenshot of Playbook add form
4

Save

Click Save. The registered Playbook takes effect from the next triage onward.
Playbook list screenshot

​
Playbook specifications

​
General-purpose Playbook

When the Triage Agent cannot match an inquiry to any other Playbook, it falls back to a general-purpose Playbook. This fallback Playbook is stored as procedures/general.md. It appears in the list like any other Playbook and can be edited or deleted.
Keep the general-purpose Playbook’s content updated to provide a reasonable fallback experience. Deleting it removes the fallback response for unmatched inquiries, so proceed with caution.

​
When changes take effect

Registered or updated Playbook content takes effect from the next triage onward. Sessions already in progress are not affected.

​
Playbook step notation

Playbook steps contain a natural language description of the action the AI agent should perform. Use the {{tool_name}} syntax in the step body to explicitly reference a tool the AI agent should call.

​
How to reference tools

Embed {{tool_name}} inline in the step text. 
### Step 1: Look up user account

Use {{mcp__admina__get_people_accounts}} to check the user's account status.
If the account is locked, proceed to Step 2. If not found, escalate.
Typing {{ in the text editor opens a tool-name autocomplete menu (typing / opens the full command menu). You can pick a tool from the menu instead of typing it by hand.

​
Available tools

Knowledge and document search
ToolCapability
{{document_search}}Searches the Knowledge Base and automatically falls back to web search when there is no hit (recommended)
{{web_search}}Search the web directly for up-to-date information
{{read_markdown_file}}Read a specific document file
Admina data lookup (MCP)
ToolCapability
{{mcp__admina__get_people_accounts}}Retrieve a user’s team, employment status, and SaaS account information
{{mcp__admina__get_devices}}Retrieve a user’s device information
{{mcp__admina__get_identities}}Retrieve a user’s identity and MFA enrollment status
{{mcp__admina__get_services}}Retrieve the list of available service integrations
{{mcp__admina__get_service_accounts}}Retrieve the account list for a specific service
Admina MCP tools require the Admina MCP integration to be enabled in Settings > Integration. Tool names follow the {{mcp__admina__tool_name}} format. Check with your admin for the full list of available Admina tools.
Core tools
ToolCapability
{{check_session_status}}Check the current status and assignee of the session (always call before {{escalation}})
{{escalation}}Escalate to an operator or team
{{escalation_contact}}Display a specific contact as a chip in the step text (see below)
{{suggest_session_complete}}Confirm resolution with the user and prompt session close
{{start_workflow}}Start tracking a named workflow (e.g. password-reset)
{{update_conversation_data}}Save information collected during the conversation
{{investigation}} and {{resolution_planning}} were removed as part of a prompt model refactor. The equivalent judgment is now built into the AI agent’s core instructions, so Playbooks no longer need to call them explicitly.

​
Contact chip syntax ({{escalation_contact}})

{{escalation_contact}} is a different tool from {{escalation}}. It isn’t an action the AI performs — it’s notation for displaying a contact clearly in the step text. Embedded with no attributes, it resolves to your tenant’s configured contact. Adding an attribute displays a specific contact as an icon-labeled chip.
SyntaxIcon shown
{{escalation_contact}}Generic icon (resolves to the tenant-configured value)
{{escalation_contact name="display name"}}Person icon
{{escalation_contact slack_name="display name"}}Slack icon
{{escalation_contact teams_username="display name"}}Microsoft Teams icon
{{escalation_contact line_id="display name"}}LINE WORKS icon
{{escalation_contact email="email address"}}Mail icon
If you have questions, check with {{escalation_contact name="Taro Yamada"}}.

​
Example

## Password Reset Procedure

### Step 1: Check user account

Use {{mcp__admina__get_people_accounts}} to verify account status.
If the account is locked, proceed to Step 2. If not found, escalate.

### Step 2: Search internal documentation

Use {{document_search}} to find the password reset procedure.
Present the found procedure to the user clearly.

### Step 3: Confirm resolution

Check whether the issue is resolved.
If resolved, use {{suggest_session_complete}} to close the session.
If not resolved, use {{check_session_status}} to check status, and if not already escalated, use {{escalation}} to hand off to the IT team.
Tool references are optional. The AI agent selects tools from context even without them, but using {{tool_name}} makes tool selection more reliable and ensures expected behavior.
Last modified on July 10, 2026