Agentic InfraOps - Agent Workflow

Agentic InfraOps - Agent Workflow

Version 3.7.8 | Changelog

This document describes the 7-step agent workflow for Azure infrastructure development with Agentic InfraOps.

How to Invoke Agents

1. Press Ctrl+Alt+I to open Copilot Chat
2. Select the agent from the agent picker dropdown at the top of the chat
3. Type your prompt and submit
4. Wait for approval before proceeding to the next step

Note: Requirements Agent

VS Code includes a built-in Plan agent for general planning tasks. This repository uses a custom Requirements agent (requirements.agent.md) specifically designed for Azure infrastructure requirements gathering. The custom agent includes Azure-specific instructions, templates, and handoffs to other agents in this 7-step workflow.

Overview

%%{init: {'theme':'neutral'}}%%
graph TB
    subgraph "Step 1: Requirements"
        P["Project Planner<br/>(custom agent)"]
    end

    subgraph "Step 2: Architecture"
        A["architect<br/>(NO CODE)"]
        MCP["💰 Azure Pricing MCP"]
    end

    subgraph "Step 3: Design Artifacts"
        D["📊 diagram<br/>(-des suffix)"]
        ADR1["📝 adr<br/>(-des suffix)"]
    end

    subgraph "Step 4: Planning"
        B["bicep-plan<br/>(governance discovery)"]
    end

    subgraph "Step 5: Implementation"
        I["bicep-code<br/>(code generation)"]
    end

    subgraph "Step 6: Deploy"
        DEP["🚀 Deploy to Azure<br/>(Deploy Agent)"]
    end

    subgraph "Step 7: As-Built Artifacts"
        D2["📊 diagram<br/>(-ab suffix)"]
        ADR2["📝 adr<br/>(-ab suffix)"]
        WL["📚 workload-documentation"]
    end

    P -->|"requirements"| A
    MCP -.->|"pricing data"| A
    A -->|"architecture"| D
    A -->|"architecture"| ADR1
    D --> B
    ADR1 --> B
    A -->|"skip artifacts"| B
    B -->|"plan"| I
    I -->|"code complete"| DEP
    DEP -->|"deployed"| D2
    DEP -->|"deployed"| ADR2
    DEP -->|"deployed"| WL

    style P fill:#e1f5fe
    style A fill:#fff3e0
    style MCP fill:#fff9c4
    style D fill:#f3e5f5
    style ADR1 fill:#e8eaf6
    style B fill:#e8f5e9
    style I fill:#fce4ec
    style DEP fill:#c8e6c9
    style D2 fill:#f3e5f5
    style ADR2 fill:#e8eaf6
    style WL fill:#e3f2fd

Workflow Steps

Step 1: Requirements (Project Planner)

Use the Project Planner custom agent to gather comprehensive requirements.

📋 Requirements Template: See .github/prompts/plan-requirements.prompt.md

Key Information to Capture:

Example Prompt:

Create deployment plan for HIPAA-compliant patient portal.
Business context: Healthcare startup serving 10,000 patients.
NFRs: 99.95% SLA, RTO=4hrs, RPO=1hr, <2s page load.
Compliance: HIPAA, SOC 2, data residency in EU.
Budget: $3,000/month maximum.

Step 2: Architecture (architect)

See agent definition: .github/agents/architect.agent.md

Validates requirements include NFRs before proceeding with WAF assessment.

Step 3: Design Artifacts (Optional)

Step 4: Governance Discovery

The bicep-plan agent now discovers Azure Policy constraints before creating the plan:

1. Queries Azure Resource Graph for policy assignments
2. Generates agent-output/{project}/04-governance-constraints.md and .json
3. Ensures implementation plan complies with org policies

Step 6: Deploy

Deploy infrastructure to Azure using generated scripts:

1. Run deploy.ps1 with -WhatIf to preview changes
2. Review deployment plan and approve
3. Execute deployment
4. Generates agent-output/{project}/06-deployment-summary.md

Step 7: As-Built Artifacts (Optional)

💰 Azure Pricing MCP Integration

The Azure Pricing MCP Server provides real-time cost data to agents during infrastructure planning. This eliminates guesswork and ensures accurate cost estimates.

Available Tools

How It Works

%%{init: {'theme':'neutral'}}%%
sequenceDiagram
    participant User
    participant Architect as architect
    participant MCP as Azure Pricing MCP
    participant Azure as Azure Retail Prices API

    User->>Architect: "Recommend App Service SKU"
    Architect->>MCP: azure_discover_skus("App Service")
    MCP->>Azure: GET /api/retail/prices?serviceName=...
    Azure-->>MCP: [P1v4: $0.282/hr, P2v4: $0.564/hr, ...]
    MCP-->>Architect: SKU options with prices
    Architect-->>User: "P1v4 recommended: $206/mo"

Setup

The MCP server is automatically configured when you open the Dev Container. Manual setup:

cd mcp/azure-pricing-mcp
python -m venv .venv
source .venv/bin/activate  # Windows: .venv\Scripts\activate
pip install -r requirements.txt

Configuration

Pre-configured in .vscode/mcp.json:

{
  "servers": {
    "azure-pricing": {
      "type": "stdio",
      "command": "${workspaceFolder}/mcp/azure-pricing-mcp/.venv/bin/python",
      "args": ["-m", "azure_pricing_mcp"],
      "cwd": "${workspaceFolder}/mcp/azure-pricing-mcp/src"
    }
  }
}

Creating Cost Documentation

During the architecture phase, ask:

• "save costs" - Creates docs/{project}-cost-estimate.md
• "save" - Creates docs/{project}-waf-assessment.md

Fallback

If MCP tools are unavailable, agents will use the MCP Pricing Fallback Chain:

1. Azure Pricing MCP (mcp_azure-pricing_*) - First choice, real-time data
2. fetch_webpage - Fetch official Azure pricing pages for specific services
3. Azure Retail API - Direct API query via curl for programmatic access
4. Azure Pricing Calculator - Manual fallback (link for user reference)
5. Web search - Last resort for general pricing information

How to Use

Starting the Workflow

1. Open GitHub Copilot Chat (Ctrl+Alt+I)
2. Click the Agent button or press Ctrl+Shift+A
3. Select @requirements to start with requirements
4. Follow the agent handoffs through each step

Approval Gates

Each step requires your explicit approval before proceeding:

Example Conversation Flow

You: @requirements Create a HIPAA-compliant patient portal with Azure App Service and SQL Database

Requirements Agent: [Generates requirements plan]
           Do you approve this plan?

You: yes

[Handoff to architect]

Architect: [Provides WAF assessment with scores]
           [Queries Azure Pricing MCP for real-time costs]

           💰 Cost Breakdown (via MCP):
           ┌─────────────────────────┬──────────────┐
           │ Resource                │ Monthly Cost │
           ├─────────────────────────┼──────────────┤
           │ App Service P1v4 (x2)   │ $412         │
           │ Azure SQL S3            │ $150         │
           │ Key Vault Standard      │ $3           │
           │ Log Analytics (5GB/day) │ $50          │
           └─────────────────────────┴──────────────┘
           Total: ~$615/month

           Do you approve? Want a diagram first?

You: yes generate the diagram

Architect: [Invokes diagram]
           ✅ Created docs/diagrams/patient-portal/architecture.py
           ✅ Generated architecture.png

           Ready to continue to Bicep planning?

You: approve

[Handoff to bicep-plan]

Planner: [Creates implementation plan in agent-output/{project}/]
         [Validates costs with Azure Pricing MCP]
         Do you approve this implementation plan?

You: yes

[Handoff to bicep-code]

Implementer: [Generates Bicep templates]
             Validation: bicep build ✅ | bicep lint ✅
             Do you approve this implementation?

You: deploy

Implementer: [Executes deployment to Azure]

Agent Responsibilities

@requirements (Custom Agent)

• Input: Natural language requirements
• Output: Structured requirements plan (01-requirements.md)
• Handoff: To architect for WAF assessment

architect

• Input: Requirements from @requirements or user
• Output: WAF pillar assessment, SKU recommendations (no cost estimates)
• Integrations:
  • 💰 Uses Azure Pricing MCP for SKU pricing context
  • 📊 Can invoke diagram for visualizations (Step 3)
  • 📝 Can invoke adr for decisions (Step 3)
• Limitations: ❌ Cannot create or edit Bicep code files
• Documentation: Creates 02-architecture-assessment.md only (cost estimates in Step 3)

diagram

• Input: Architecture context from architect OR bicep-code
• Output: Python diagram file in agent-output/{project}/ + PNG image
• Artifact Suffix:
  • -des when called from Step 3 (design phase)
  • -ab when called from Step 7 (as-built phase)
• Triggered By: User request ("generate diagram", "create as-built diagram")
• Library: Uses Python diagrams library

bicep-plan

• Input: Architecture assessment from architect
• Output:
  • Implementation plan in agent-output/{project}/04-implementation-plan.md
  • Governance constraints in agent-output/{project}/04-governance-constraints.md and .json
• Governance Discovery: Queries Azure Policy via MCP before planning
• Integrations: 💰 Uses Azure Pricing MCP for SKU cost validation
• Limitations: ❌ Cannot create actual Bicep code
• Focus: Detailed planning with AVM modules + policy compliance

bicep-code

• Input: Implementation plan from agent-output/{project}/04-implementation-plan.md
• Output:
  • Bicep templates in infra/bicep/{project}/
  • Reference file: agent-output/{project}/05-implementation-reference.md
• Limitations: Must follow the approved plan
• Validation: Runs bicep build and bicep lint
• Focus: Code generation and deployment
• Handoffs: Can invoke diagram and adr for as-built artifacts

adr (Optional)

• Input: Any architectural decision during the workflow
• Output: ADR in agent-output/{project}/
• Artifact Suffix:
  • -des when documenting design decisions (Step 3)
  • -ab when documenting implementation decisions (Step 7)
• When to Use: Major technology choices, trade-off decisions, policy exceptions

docs (Optional)

• Input: All previous artifacts (WAF assessment, plan, Bicep code)
• Output: Customer-deliverable documentation in agent-output/{project}/07-*.md
• Includes: Design document, operations runbook, resource inventory, compliance matrix
• When to Use: Customer handoffs, consulting engagements, formal documentation needs

File Locations

Troubleshooting

Agent Not Available

1. Press Ctrl+Shift+A to open agent picker
2. Verify agent files exist in .github/agents/
3. Restart VS Code if agents don't appear

MCP Pricing Not Working

1. Check MCP server status: Command Palette → "MCP: List Servers"
2. Verify azure-pricing shows 6 tools
3. If not, run setup: cd mcp/azure-pricing-mcp && pip install -r requirements.txt
4. Restart VS Code

Handoff Not Working

1. Look for the handoff button in chat response
2. If no button, manually switch agents with Ctrl+Shift+A
3. Copy relevant context to new agent conversation

Validation Failures

Related Documentation

• Time Savings Evidence — Quantified efficiency gains (45 min vs 18+ hours)
• Copilot Portfolio Showcase — Real-world projects
• ADR-003: AVM-First Approach — Why we use Azure Verified Modules
• ADR-004: Region Defaults — Default swedencentral, fallback guidance
• Copilot Instructions
• Azure Pricing MCP README
• README