Research Components

After architecture planning, research each selected component to gather best practices before generating artifacts.

Process

1. Identify Components — List all Azure services from architecture plan
2. Load Service References — For each service, load services/<service>/README.md first, then specific references as needed
3. Check Resource Naming Rules — For each resource type, check resource naming rules for valid characters, length limits, and uniqueness scopes
4. Load Recipe References — Load the selected recipe's guide (e.g., AZD) and its IAC rules, MCP best practices, and schema tools listed in its "Before Generation" table
5. Check Region Availability — Verify all selected services are available in the target region per region-availability.md
6. Check Provisioning Limits — Invoke azure-quotas skill to validate that the selected subscription and region have sufficient quota/capacity for all planned resources. Complete Step 6 of the plan template in two phases: (1) prepare resource inventory with deployment quantities, (2) fetch quotas and validate capacity using azure-quotas skill
7. Load Runtime References — For containerized apps, load language-specific production settings (e.g., Node.js)
8. Invoke Related Skills — For deeper guidance, invoke mapped skills from the table below
9. Document Findings — Record key insights in .azure/deployment-plan.md

Service-to-Reference Mapping

Research Instructions

Step 1: Load Internal References (Progressive Loading)

For each selected service, load the README.md first, then load specific files as needed:

Selected: Container Apps, Cosmos DB, Key Vault

→ Load: services/container-apps/README.md (overview)
  → If need Bicep: services/container-apps/bicep.md
  → If need Terraform: services/container-apps/terraform.md
  → If need scaling: services/container-apps/scaling.md
  → If need health probes: services/container-apps/health-probes.md

→ Load: services/cosmos-db/README.md (overview)
  → If need partitioning: services/cosmos-db/partitioning.md
  → If need SDK: services/cosmos-db/sdk.md

→ Load: services/key-vault/README.md (overview)
  → If need SDK: services/key-vault/sdk.md

Step 2: Invoke Related Skills (When Deeper Guidance Needed)

Invoke related skills for specialized scenarios:

Skill/Reference Invocation Pattern:

For Azure Functions:

1. Load: selection.md (decision tree)
2. Follow: composition.md (algorithm)
3. Result: Base template + recipe composition (never synthesize IaC)

For PostgreSQL:

1. Handle passwordless auth patterns directly without a separate skill

Step 3: Document in Plan

Add research findings to .azure/deployment-plan.md under a ## Research Summary section with source references and key insights per component.

Common Research Patterns

Web Application + API + Database (Cosmos DB)

1. Load: services/container-apps/README.md → bicep.md or terraform.md, scaling.md
2. Load: services/cosmos-db/README.md → partitioning.md
3. Load: services/key-vault/README.md
4. Invoke: azure-observability (monitoring setup)
5. Review service-specific security guidance directly before generation

Container Apps + API + SQL Database

1. Load: services/container-apps/README.md → bicep.md or terraform.md, scaling.md
2. Load: services/sql-database/README.md → bicep.md, auth.md
3. Load: services/key-vault/README.md
4. Review auth.md directly for Entra-only auth configuration

App Service + API + SQL Database

1. Load: services/app-service/README.md → bicep.md
2. Load: services/sql-database/README.md → bicep.md, auth.md
3. Load: services/key-vault/README.md
4. Review auth.md directly for Entra-only auth configuration

Serverless Event-Driven

1. Load: services/functions/README.md (contains mandatory composition workflow)
2. Load: services/event-grid/README.md or services/service-bus/README.md (if using messaging)
3. Load: services/storage/README.md (if using queues/blobs)
4. Invoke: azure-observability (distributed tracing)

AI Application

1. Invoke: microsoft-foundry (AI patterns and best practices)
2. Load: services/container-apps/README.md → bicep.md or terraform.md
3. Load: services/cosmos-db/README.md → partitioning.md (vector storage)
4. Review Key Vault and Foundry references directly for API key management

GitHub Copilot SDK Application

1. Invoke: azure-hosted-copilot-sdk skill (scaffold, infra, model config)
2. After it completes, resume azure-prepare workflow (validate → deploy)

After Research

Proceed to Generate Artifacts step with research findings applied.