feat(python): Add Copilot Studio sample agent

[Yogeshp-MSFT]

Summary

Adds a new Python sample that bridges Microsoft Copilot Studio low-code agents into the Agent 365 managed environment. Messages arriving through Agent 365 channels (Teams, email) are forwarded to a published Copilot Studio agent via CopilotClient, and responses are relayed back — giving low-code agents access to enterprise identity, notifications, and observability.

What's included

python/copilot-studio/sample-agent/
├── main.py                 # Server entry point — CopilotStudioAgentHost, aiohttp, observability
├── agent.py                # MyAgent — message/notification routing
├── client.py               # McsClient — CopilotClient wrapper with InferenceScope
├── .env.template           # Environment variable template
├── .gitignore              # Python + A365 ignores
├── pyproject.toml          # Dependencies (hatchling)
├── requirements.txt        # For Azure Oryx build
├── ToolingManifest.json    # Empty (proxies to Copilot Studio, no direct MCP tools)
├── README.md               # Comprehensive docs
└── docs/
    └── design.md           # Architecture & design document

Patterns demonstrated

Pattern	Implementation
Copilot Studio integration	microsoft-agents-copilotstudio-client — CopilotClient + ConnectionSettings for Direct Line connectivity
Agentic auth (Power Platform)	OBO token exchange for https://api.powerplatform.com/.default via the AGENTIC auth handler
Observability	AgenticTokenCache pattern — tokens acquired at runtime from TurnContext, not static env vars. InferenceScope wraps every Copilot Studio call
Notifications	Email notification handling via AgentNotification + EmailResponse
Multiple messages	Immediate ack + typing indicator loop + final response
User identity	Logs activity.from_property fields on every turn
Install/uninstall	Handles installationUpdate activity

Key design decisions

• No A365_AGENT_APP_INSTANCE_ID / A365_AGENTIC_USER_ID env vars — Observability tokens are acquired from the incoming TurnContext using the AgenticTokenCache pattern per the Microsoft OpenTelemetry Distro docs
• AUTH_HANDLER_NAME=AGENTIC (uppercase) — must match the handler key in AGENTAPPLICATION__USERAUTHORIZATION__HANDLERS__AGENTIC__... (Python dict lookup is case-sensitive)
• CopilotStudio.Copilots.Invoke permission required on the blueprint — configurable via a365 setup permissions copilotstudio
• Root-level file layout (no src/ package) — matches the google-adk sample convention and simplifies Azure App Service deployment

How to test

cd python/copilot-studio/sample-agent
uv venv && uv sync
cp .env.template .env
# Fill in Copilot Studio Environment ID + Agent Identifier
# Fill in service connection credentials from a365 setup all
python main.py

Tested end-to-end with:

• Agents Playground (anonymous mode)
• Teams via devtunnel (agentic auth)
• Azure App Service deployment (zip deploy via az webapp deploy)
• Observability export verified: HTTP 200 success, rejectedSpans: 0

Dependencies

• microsoft-agents-copilotstudio-client — Copilot Studio connectivity
• microsoft-agents-hosting-aiohttp / microsoft-agents-hosting-core — M365 Agents SDK hosting
• microsoft-agents-a365-notifications — Email/Teams notifications
• microsoft-agents-a365-runtime — Observability token helpers
• microsoft-opentelemetry — OpenTelemetry distro (InferenceScope, BaggageBuilder, AgenticTokenCache)

Related

• Copilot Studio docs: https://learn.microsoft.com/en-us/microsoft-copilot-studio/
• OpenTelemetry Distro docs: https://learn.microsoft.com/en-us/microsoft-agent-365/developer/microsoft-opentelemetry?tabs=python

---

[Yogeshp-MSFT]

Hi @biswapm can you review this pr ?