19 KiB
| description | model | tools | |
|---|---|---|---|
| Best practices and guidelines for developing declarative agents for Microsoft 365 Copilot following schema v1.5 specifications with Microsoft 365 Agents Toolkit integration | GPT-4.1 |
|
Microsoft 365 Declarative Agents Development Guidelines
Comprehensive instructions for developing high-quality declarative agents for Microsoft 365 Copilot using schema v1.5 with Microsoft 365 Agents Toolkit integration.
Core Development Principles
Schema v1.5 Compliance
Always adhere to the official Microsoft declarative agent schema v1.5:
{
"$schema": "https://raw.githubusercontent.com/microsoft/copilot-studio-schemas/main/schemas/v1.5/declarative-copilot.schema.json",
"version": "v1.0",
"name": "[Agent Name]",
"description": "[Agent Description]",
"instructions": "[Detailed Instructions]",
"capabilities": [],
"conversation_starters": []
}
Character Limit Constraints
Strictly enforce character limits:
- name: Maximum 100 characters
- description: Maximum 1000 characters
- instructions: Maximum 8000 characters
- conversation_starters.text: Maximum 100 characters each
Microsoft 365 Agents Toolkit Integration
Leverage the Microsoft 365 Agents Toolkit (teamsdevapp.ms-teams-vscode-extension) for:
- TypeScript/TypeSpec development with type safety
- Local testing with Agents Playground
- Environment management (dev/staging/production)
- Performance monitoring and optimization
- Automated validation and deployment
Capability Configuration Guidelines
1. WebSearch Capability
Use for real-time information retrieval and current data access:
{
"type": "WebSearch",
"configuration": {
"enabled": true,
"search_domains": ["specific-domains-if-needed"],
"result_limit": 10,
"content_filtering": {
"safe_search": "strict",
"relevance_threshold": 0.7
}
}
}
Best Practices:
- Enable only when real-time data is essential
- Configure domain restrictions for enterprise security
- Set appropriate result limits to manage token usage
- Use content filtering for compliance requirements
2. OneDriveAndSharePoint Capability
Configure for file operations and content management:
{
"type": "OneDriveAndSharePoint",
"configuration": {
"scopes": [
"https://graph.microsoft.com/Files.Read",
"https://graph.microsoft.com/Files.ReadWrite"
],
"file_types": [".docx", ".xlsx", ".pptx", ".pdf"],
"size_limits": {
"max_file_size": "10MB",
"max_concurrent_files": 5
}
}
}
Best Practices:
- Request minimal necessary scopes
- Specify supported file types explicitly
- Set reasonable size limits for performance
- Implement proper error handling for file operations
3. GraphConnectors Capability
Integrate custom data sources through Graph Connectors:
{
"type": "GraphConnectors",
"configuration": {
"connection_id": "your-connector-id",
"entity_types": ["CustomDocument", "KnowledgeArticle"],
"query_capabilities": {
"semantic_search": true,
"faceted_search": true,
"result_ranking": "relevance"
}
}
}
Best Practices:
- Ensure connector is properly configured and indexed
- Define specific entity types for targeted search
- Enable semantic search for better user experience
- Configure result ranking based on business needs
4. Function Capability
Implement custom business logic and API integrations:
{
"type": "Function",
"function_definition": {
"name": "analyze_sales_data",
"description": "Analyze sales performance metrics and generate insights",
"parameters": {
"type": "object",
"properties": {
"time_period": {
"type": "string",
"description": "Analysis time period (e.g., 'last_quarter', 'ytd')",
"enum": ["last_month", "last_quarter", "ytd", "custom"]
},
"metrics": {
"type": "array",
"items": {
"type": "string",
"enum": ["revenue", "units_sold", "conversion_rate", "customer_acquisition"]
},
"description": "Specific metrics to analyze"
}
},
"required": ["time_period", "metrics"]
}
}
}
Best Practices:
- Use clear, descriptive function names
- Provide comprehensive parameter descriptions
- Define appropriate parameter types and constraints
- Include required field specifications
- Implement robust error handling in function implementation
5. MicrosoftGraph Capability
Access Microsoft 365 services and data:
{
"type": "MicrosoftGraph",
"configuration": {
"scopes": [
"https://graph.microsoft.com/User.Read",
"https://graph.microsoft.com/Calendars.ReadWrite",
"https://graph.microsoft.com/Mail.Read"
],
"delegated_permissions": true,
"api_version": "v1.0"
}
}
Best Practices:
- Request only necessary scopes for functionality
- Use delegated permissions for user context
- Specify stable API versions (v1.0) for production
- Handle authentication errors gracefully
TypeSpec Development Best Practices
Type-Safe Agent Definitions
Use TypeSpec for type-safe agent development:
import "@typespec/http";
import "@microsoft/declarative-agent-manifest";
using Microsoft.DeclarativeAgent;
/** Business Intelligence Agent for sales data analysis */
@DeclarativeAgent
model BusinessIntelligenceAgent {
/** Agent display name */
@maxLength(100)
name: "Sales Analytics Assistant";
/** Agent description for users */
@maxLength(1000)
description: "AI-powered assistant for sales data analysis, reporting, and insights generation";
/** Detailed agent instructions */
@maxLength(8000)
instructions: `You are a specialized business intelligence assistant focused on sales data analysis...`;
/** Agent capabilities */
capabilities: [
FunctionCapability,
MicrosoftGraphCapability
];
/** Pre-defined conversation starters */
conversation_starters: ConversationStarter[];
}
model FunctionCapability extends Capability {
type: "Function";
function_definition: SalesAnalysisFunction;
}
model SalesAnalysisFunction {
name: "analyze_sales_data";
description: "Analyze sales performance data and generate insights";
parameters: SalesAnalysisParameters;
}
Compilation and Validation
Compile TypeSpec to JSON and validate:
# Compile TypeSpec to JSON manifest
tsp compile models/agent.tsp --emit @microsoft/declarative-agent-manifest
# Validate against schema
teamsapp validate --manifest-path ./dist/manifest.json
Agent Instruction Engineering
Instruction Structure Template
Follow this structure for comprehensive agent instructions:
## Role Definition
You are [specific role] specialized in [domain expertise].
## Core Responsibilities
- [Primary responsibility 1]
- [Primary responsibility 2]
- [Primary responsibility 3]
## Behavioral Guidelines
- Maintain [specific behavior trait]
- Always [specific action requirement]
- Never [specific prohibition]
## Response Format
When responding to user queries:
1. [Step 1 requirement]
2. [Step 2 requirement]
3. [Step 3 requirement]
## Capability Usage
- Use WebSearch for [specific scenarios]
- Use Function capabilities for [specific scenarios]
- Use MicrosoftGraph for [specific scenarios]
## Error Handling
If you encounter issues:
- [Error scenario 1]: [Response strategy]
- [Error scenario 2]: [Response strategy]
## Limitations
- I cannot [specific limitation 1]
- I do not have access to [specific limitation 2]
Token Efficiency Optimization
Optimize instructions for token efficiency:
- Use concise language: Avoid redundant phrases
- Prioritize information: Place most important guidance first
- Use structured formatting: Bullet points and numbered lists
- Avoid repetition: Each instruction should provide unique value
- Test token usage: Monitor actual token consumption in testing
Conversation Starter Best Practices
Effective Conversation Starters
Design conversation starters that demonstrate agent capabilities:
{
"conversation_starters": [
{
"text": "Analyze last quarter's sales performance by region"
},
{
"text": "Generate a monthly revenue report with trend analysis"
},
{
"text": "Compare this year's sales metrics to last year"
},
{
"text": "Identify top-performing products and growth opportunities"
}
]
}
Guidelines:
- Specific and actionable: Each starter should trigger a specific workflow
- Capability demonstration: Show what the agent can do
- User language: Use terminology familiar to target users
- Varied complexity: Include simple and complex examples
- Business value: Focus on outcomes users care about
Microsoft 365 Agents Toolkit Workflow
Development Environment Setup
-
Install Prerequisites:
# Install Teams Toolkit CLI npm install -g @microsoft/teamsapp-cli # Verify installation teamsapp --version -
VS Code Extension Setup:
- Install Microsoft 365 Agents Toolkit extension
- Configure workspace settings for declarative agents
- Set up debugging configuration
-
Project Initialization:
# Create new declarative agent project teamsapp new declarative-agent --name "MyBusinessAgent" # Initialize TypeSpec development npm install @typespec/compiler @microsoft/declarative-agent-manifest
Local Development and Testing
-
Agents Playground Integration:
- Load manifest in toolkit
- Test conversation flows interactively
- Validate capability responses
- Monitor token usage and performance
-
Environment Configuration:
{ "environments": { "development": { "manifest_path": "./manifests/dev-manifest.json", "test_scenarios": "./test-data/dev-scenarios.json", "mock_data": true }, "staging": { "manifest_path": "./manifests/staging-manifest.json", "test_scenarios": "./test-data/staging-scenarios.json", "integration_testing": true }, "production": { "manifest_path": "./manifests/prod-manifest.json", "monitoring_enabled": true, "performance_tracking": true } } } -
Debugging and Troubleshooting:
interface DebuggingConfig { log_level: "debug" | "info" | "warn" | "error"; trace_capability_calls: boolean; monitor_token_usage: boolean; validate_responses: boolean; }
Performance Optimization
-
Token Usage Monitoring:
interface TokenUsageMetrics { instruction_tokens: number; capability_tokens: number; response_tokens: number; total_tokens: number; optimization_suggestions: string[]; } -
Response Time Optimization:
- Capability ordering optimization
- Parallel capability execution where possible
- Caching strategies for repeated queries
- Connection pooling for external APIs
-
Memory Management:
- Conversation context optimization
- State management best practices
- Resource cleanup procedures
Security and Compliance Guidelines
Authentication and Authorization
-
OAuth 2.0 Implementation:
{ "authentication": { "type": "OAuth2", "provider": "Microsoft", "scopes": [ "https://graph.microsoft.com/User.Read", "https://graph.microsoft.com/Files.Read" ], "token_management": { "refresh_strategy": "automatic", "expiration_handling": "graceful_degradation" } } } -
Scope Minimization:
- Request only necessary permissions
- Document scope justification
- Implement permission checking
- Handle scope denial gracefully
Data Protection
-
PII Handling:
interface PIIHandlingConfig { detection_enabled: boolean; redaction_strategy: "mask" | "remove" | "encrypt"; audit_logging: boolean; retention_policy: string; } -
Data Encryption:
- Encrypt data in transit (TLS 1.3)
- Encrypt sensitive data at rest
- Implement key rotation policies
- Use Azure Key Vault for secrets
Compliance Framework
-
Audit Logging:
{ "audit_config": { "log_user_interactions": true, "log_capability_usage": true, "log_data_access": true, "retention_period": "7_years", "compliance_frameworks": ["SOC2", "GDPR", "ISO27001"] } } -
Compliance Validation:
- Regular security assessments
- Compliance monitoring
- Incident response procedures
- Data breach notification protocols
Testing and Validation Procedures
Unit Testing
Test individual components:
describe('Sales Analysis Agent', () => {
test('validates manifest schema compliance', () => {
const manifest = loadManifest('./dist/manifest.json');
expect(validateSchema(manifest)).toBe(true);
});
test('respects character limits', () => {
const manifest = loadManifest('./dist/manifest.json');
expect(manifest.name.length).toBeLessThanOrEqual(100);
expect(manifest.description.length).toBeLessThanOrEqual(1000);
expect(manifest.instructions.length).toBeLessThanOrEqual(8000);
});
test('validates capability configurations', () => {
const manifest = loadManifest('./dist/manifest.json');
manifest.capabilities.forEach(capability => {
expect(validateCapabilityConfig(capability)).toBe(true);
});
});
});
Integration Testing
Test capability interactions:
describe('Capability Integration', () => {
test('function capability responds correctly', async () => {
const response = await testCapability('Function', {
function_name: 'analyze_sales_data',
parameters: { time_period: 'last_quarter', metrics: ['revenue'] }
});
expect(response.success).toBe(true);
expect(response.data).toBeDefined();
});
test('graph capability handles authentication', async () => {
const response = await testCapability('MicrosoftGraph', {
endpoint: '/me',
method: 'GET'
});
expect(response.authenticated).toBe(true);
});
});
Performance Testing
Validate performance requirements:
describe('Performance Requirements', () => {
test('response time under 3 seconds', async () => {
const startTime = Date.now();
const response = await sendMessage('Analyze sales data');
const responseTime = Date.now() - startTime;
expect(responseTime).toBeLessThan(3000);
});
test('token usage within limits', async () => {
const response = await sendMessage('Generate quarterly report');
expect(response.token_usage.total).toBeLessThan(4000);
});
});
Production Deployment Guidelines
Pre-Deployment Checklist
- Schema validation passes
- Character limits respected
- All capabilities tested individually
- Integration testing completed
- Performance benchmarks met
- Security review completed
- Compliance validation finished
- Documentation complete
- Monitoring configured
- Rollback plan prepared
Deployment Pipeline
# .github/workflows/deploy-agent.yml
name: Deploy Declarative Agent
on:
push:
branches: [main]
jobs:
validate:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v3
- name: Setup Node.js
uses: actions/setup-node@v3
with:
node-version: '18'
- name: Install dependencies
run: npm install
- name: Compile TypeSpec
run: tsp compile models/agent.tsp
- name: Validate manifest
run: teamsapp validate --manifest-path ./dist/manifest.json
- name: Run tests
run: npm test
deploy:
needs: validate
runs-on: ubuntu-latest
steps:
- name: Deploy to staging
run: teamsapp deploy --env staging
- name: Run integration tests
run: npm run test:integration
- name: Deploy to production
run: teamsapp deploy --env production
Monitoring and Maintenance
-
Performance Monitoring:
{ "monitoring": { "response_time_tracking": true, "token_usage_monitoring": true, "error_rate_tracking": true, "user_satisfaction_metrics": true } } -
Incident Response:
- Automated alerting for failures
- Escalation procedures
- Rollback capabilities
- Post-incident reviews
-
Continuous Improvement:
- Regular performance reviews
- User feedback integration
- Capability optimization
- Security updates
Common Patterns and Anti-Patterns
Design Patterns ✅
-
Single Responsibility Pattern:
{ "name": "Sales Analytics Assistant", "focus": "sales_data_analysis", "capabilities": ["Function", "MicrosoftGraph"] } -
Graceful Degradation Pattern:
if (primaryCapabilityFails) { return fallbackResponse(); } -
Progressive Enhancement Pattern:
{ "basic_functionality": ["simple_queries"], "enhanced_functionality": ["complex_analysis", "visualizations"] }
Anti-Patterns ❌
-
Kitchen Sink Anti-Pattern:
// Don't do this - too many unrelated capabilities { "capabilities": [ "WebSearch", "OneDriveAndSharePoint", "GraphConnectors", "Function", "MicrosoftGraph" ] } -
Overly Generic Instructions:
// Don't do this - too vague "You are a helpful assistant that can do many things." -
Character Limit Violations:
// Don't do this - exceeds limits { "name": "Super Long Agent Name That Exceeds The Maximum Character Limit Of One Hundred Characters And Should Be Shortened", "description": "[description over 1000 characters]" }
Follow these comprehensive guidelines to create high-quality, performant, and secure declarative agents for Microsoft 365 Copilot that provide exceptional user experiences.