Prompt Administration

Release Notes

1.1

• Published from Beta

Prior 1.1

• Beta Testing

Overview

The Prompt Management Metablock provides a comprehensive system for creating, editing, and maintaining AI prompts within the XMPro Application Designer. This Metablock enables organizations to manage prompt libraries with version control, ensuring consistent AI interactions across applications. The Metablock supports internationalization, role-based access control, and maintains a complete audit trail of all prompt modifications.

The Metablock utilizes Neo4j for data persistence and implements efficient batch loading patterns to handle large prompt libraries. Its responsive interface adapts to different cultural preferences, including right-to-left languages, while providing consistent functionality across modern browsers and offering both light and dark themes.

Dependencies

• Neo4j Browser Driver v5.17.0
• Modern browser with ES6+ support

Installation and Setup

• Add a Metablock component to your page in the App Designer
• Open the Block Properties for the Metablock
• Upload the following files in the Block Properties:

• Configure the Value Mappings:

• Save and publish the page

Key Features

• Comprehensive prompt CRUD operations
• Version control with activation management
• Audit history tracking
• Tag-based organization
• Role-based access control
• Internationalization support
• Dark/Light theme modes
• Real-time search and filtering
• Batch loading with infinite scroll
• Import/Export functionality

Security Considerations

Access Control

Four access levels with increasing privileges:

• User: Personal prompts only
• Restricted: Department-level access
• Admin: Organization-wide access
• System: Full system access

Data Protection

• Secure Neo4j connections
• Role-based data access
• Audit trail for all changes
• Version control for recovery

System Prompt Id Structure

Structure: XMAGS-[Process]-PROMPT-[Version]

Example: XMAGS-OBS-PROMPT-001

Components:

• XMAGS: Fixed identifier for the Multi-Agent System
• PROMPT: Fixed identifier for Prompt type

Examples

Fields

Prompt Management Guide

User Interface

The UI is divided into three main sections:

1. Prompt List: Displays all prompts, organized by categories.
2. Prompt Editor: Form for creating or editing prompts.
3. Metadata Panel: The metadata for the prompt selected.

Creating a New Prompt

1. Click the “New Prompt” button in the top-left corner
2. Enter the Area and Function codes (e.g., WTR for Water, QUAL for Quality)
3. Click “Generate” to create a unique Prompt ID
4. Complete the required fields:
   • Name: Descriptive name for the prompt
   • Internal Name: System identifier (must be unique)
   • Description: Detailed explanation of prompt’s purpose
   • Model Provider: Select from available AI providers
   • Model Name: Specify the model
   • Max Tokens: Set maximum token limit
5. Enter the prompt content in the main text area
6. Add relevant tags to categorize the prompt
7. Click the floating save button to create the prompt

Editing Existing Prompts

1. Select the prompt from the list on the left
2. Review and modify desired fields in the prompt editor:
3. Click the floating save button to create a new version
4. Verify successful update in the version history

Clone Prompt

1. Select an existing prompt from the list
2. Click the “Clone Prompt” button
3. Review and modify the cloned prompt as needed
4. Click the floating save button to create and save the cloned prompt

Managing Prompt Versions

1. Select the prompt to manage
2. Access version controls in the metadata panel:
   • Use version dropdown to view available versions
   • Select specific version to review content
   • Click activate button (checkmark) to set active version
3. Review version history:
   • Click three-dot menu on prompt
   • Select “View History”
   • Review audit trail in modal window
4. Monitor active version status in prompt list

Importing Prompts

1. Click the import icon (upload arrow) in the toolbar
2. Select a JSON file containing prompt data
3. Review and modify the imported prompt as needed
4. Click the floating save button to create and save the imported prompt

Exporting Prompts

1. Select the prompt to export
2. Click the export icon (download arrow)
3. Choose save location in file browser

Search and Filtering

1. Use the search bar for real-time search:
   • Enter prompt name or ID
   • Type keywords for content search
   • Search by tag or category
2. Apply filters as needed:
   • Toggle status filters for active/inactive prompts
   • Select category filters
   • Combine multiple filters for refined results
3. Clear filters:
   • Click clear button (x) in search bar
   • Deselect active filters
   • Reset to view all prompts

Audit History

• The system maintains an audit log when an entire prompt is made inactive or reactivated.
• Users can view the audit history for each prompt.

Audit Entry Properties

Each AuditEntry node contains the following information:

• change: Description of the action or change made (e.g., “Deactivated prompt”)
• user: The user who performed the action
• timestamp: The date and time when the action occurred

Example Audit Log Entries

Here are some example audit log entries from the system:

1. Deactivation of a prompt:
   • Change: “Deactivated prompt”
   • User: “Sam Johnson”
   • Timestamp: 2024-09-01 15:04:35.284
2. Earlier deactivation of the same prompt:
   • Change: “Deactivated prompt”
   • User: “Sam Johnson”
   • Timestamp: 2024-09-01 15:02:30.304
3. Initial deactivation of the prompt:
   • Change: “Deactivated prompt”
   • User: “Sam Johnson”
   • Timestamp: 2024-09-01 15:01:21.827

Import/Export Format Examples

The following examples demonstrate the JSON structure required for importing prompts. Each example showcases different types of prompts with their specific configurations.

Basic Analysis Prompt:

{
  "prompt_id": "WTR-QUAL-PROMPT-001",
  "name": "Water Quality Analysis",
  "internal_name": "water_quality_analysis_base",
  "description": "Analyzes water quality parameters and provides detailed recommendations based on standardized water quality metrics. Suitable for both drinking water and environmental water quality assessment.",
  "model_provider": "Anthropic",
  "model_name": "claude-3-opus-20240229",
  "max_tokens": 4096,
  "prompt": "You are a water quality expert analyzing test results. Given the following water quality parameters {parameters}, analyze each parameter against standard safety thresholds. For any parameters outside acceptable ranges, explain the potential implications and provide specific recommendations for remediation. Structure your response with:\n\n1. Overall Water Quality Assessment\n2. Detailed Parameter Analysis\n3. Health and Environmental Implications\n4. Recommended Actions\n\nInclude relevant regulatory standards in your analysis where applicable.",
  "category": "Environmental",
  "type": "analysis",
  "access_level": "user",
  "tags": ["water quality", "analysis", "recommendations", "environmental"],
  "active": true,
  "version": "1",
  "author": "system.admin@organization.com"
}

System Configuration Prompt:

{
  "prompt_id": "SYS-CONF-PROMPT-001",
  "name": "System Configuration Assistant",
  "internal_name": "system_config_assistant",
  "description": "Provides guidance for system configuration tasks with a focus on security best practices and performance optimization.",
  "model_provider": "Anthropic",
  "model_name": "claude-3-opus-20240229",
  "max_tokens": 2048,
  "prompt": "As a system configuration expert, assist with the following configuration task for {system_name}. Consider these key factors:\n\n- Current Configuration: {current_config}\n- Security Requirements: {security_reqs}\n- Performance Targets: {performance_targets}\n\nProvide specific configuration recommendations that:\n1. Meet all security requirements\n2. Optimize system performance\n3. Follow industry best practices\n\nInclude both configuration values and explanations for each recommendation.",
  "category": "System",
  "type": "configuration",
  "access_level": "admin",
  "tags": ["system", "configuration", "security", "performance"],
  "active": true,
  "version": "1",
  "author": "system.admin@organization.com"
}

Data Transformation Prompt:

{
  "prompt_id": "DAT-TRAN-PROMPT-001",
  "name": "Data Format Transformer",
  "internal_name": "data_transformer_base",
  "description": "Transforms data between different formats while preserving data integrity and maintaining proper structure.",
  "model_provider": "Anthropic",
  "model_name": "claude-3-opus-20240229",
  "max_tokens": 3072,
  "prompt": "Transform the provided {source_format} data into {target_format} format. Apply the following rules:\n\n1. Maintain data integrity for all fields\n2. Follow the schema: {schema_definition}\n3. Apply data type validations: {validation_rules}\n\nInput Data:\n{input_data}\n\nProvide the transformed data along with a verification summary including any warnings or potential data quality issues.",
  "category": "Data Processing",
  "type": "transformation",
  "access_level": "restricted",
  "tags": ["data", "transformation", "validation"],
  "active": true,
  "version": "1",
  "author": "data.engineer@organization.com"
}

Data Flow

graph TD
    A[User Interface] -->|Query Prompts| B[Neo4j Database]
    B -->|Return Filtered Prompts| A
    A -->|Display Prompts| C[User]
    C -->|Create/Edit Prompt| A
    A -->|Validate Data| D[Data Validation]
    D -->|If Valid| E[Create/Update Prompt Node]
    E -->|Link Versions| F[Previous Version Node]
    E --> G[Deactivate/Reactivate Entire Prompt]
    G -->|Create Audit Entry| H[Audit Log]
    H -->|Link to Prompt| E

Prompt Management Taxonomies

Access Levels

The Prompt Library implements a role-based access control system with four levels: system, admin, restricted, and user. Each level has different permissions and capabilities within the system.

Access Level Permissions Table

Category Types

Prompt Types

Root Node

There is a root node called Library (type: prompt) to which all prompts are linked. The reasons for this are:

1. Organization: It provides a clear organizational structure, making it easier to manage and understand the relationship between prompts.

2. Querying: While it adds one more hop in some queries, it makes it much easier to perform operations on all prompts (e.g., finding all prompts, updating properties across all prompts).

3. Consistency: Having a root node allows you to enforce consistent properties or relationships across all prompts more easily.

4. Scalability: As your prompt library grows, having a root node will make it easier to manage and maintain the structure.

5. Flexibility: You can still easily query individual prompts, and the added structure doesn’t significantly impact performance for most use cases.

Querying Examples

These example Cyphers can be run on the graph database to return the specific details:

• Latest Versions Only:

   MATCH (p:Prompt)
   WHERE p.active = true
   RETURN p
   ORDER BY p.prompt_id

• Full Version of history:

Cypher MATCH path = (p:Prompt {internal_name: "planning_decision_prompt"})<-[:PREVIOUS_VERSION*]-(older:Prompt) RETURN path

Replace internal_name with the prompt in question to query.