Python: Add Container Apps Dynamic Sessions sample

python/samples/02-agents/tools/README.md

@@ -1,5 +1,8 @@
 # Tools
 
+[Container Apps Dynamic Sessions](./container_apps_dynamic_sessions.py) shows a
+function tool that executes Python in a cloud session with per-run session ownership.
+
 Samples that show how to define, configure, and control function tools for an
 agent — from basic declarations to approvals, invocation limits, session
 injection, and dynamic (progressive) tool exposure.

python/samples/02-agents/tools/container_apps_dynamic_sessions.py

@@ -0,0 +1,92 @@
+# Copyright (c) Microsoft. All rights reserved.
+
+"""Run agent-generated Python in an Azure Container Apps Dynamic Sessions pool.
+
+Prerequisites:
+- Install agent-framework-foundry, azure-identity, and httpx; sign in with az login.
+- Set FOUNDRY_PROJECT_ENDPOINT and FOUNDRY_MODEL for your model deployment.
+- Set POOL_MANAGEMENT_ENDPOINT to the Python code-interpreter pool's management
+  endpoint (az containerapp sessionpool show --query properties.poolManagementEndpoint).
+- Give the signed-in identity Azure ContainerApps Session Executor on the pool,
+  and permission to invoke the Foundry model.
+
+Each program run generates a fresh session identifier, reused for its tool calls.
+The identifier is owned by the application, never selected by the model. In a
+multi-user host, create a separate agent/tool/session binding for each authorized
+conversation. Pool cooldown controls cleanup after the last use; this sample does
+not provision or delete Azure resources. Model and session-pool usage incur costs.
+
+The tool sends code to the configured cloud sandbox; it does not execute code on
+this machine. Configure the pool's network access for your workload and do not
+pass application credentials or confidential data into generated code.
+
+REST contract:
+https://learn.microsoft.com/rest/api/data-plane/containerapps/code-execution/execute
+"""
+
+import asyncio
+import os
+from typing import Annotated, Any
+from urllib.parse import urlsplit
+from uuid import uuid4
+
+import httpx
+from agent_framework import Agent, tool
+from agent_framework.foundry import FoundryChatClient
+from azure.identity.aio import AzureCliCredential
+from dotenv import load_dotenv
+
+load_dotenv()
+
+
+async def main() -> None:
+    pool_endpoint = os.environ["POOL_MANAGEMENT_ENDPOINT"].rstrip("/")
+    endpoint = urlsplit(pool_endpoint)
+    if (
+        endpoint.scheme != "https"
+        or not endpoint.netloc
+        or "?" in pool_endpoint
+        or "#" in pool_endpoint
+        or endpoint.username is not None
+    ):
+        raise ValueError("POOL_MANAGEMENT_ENDPOINT must be a trusted HTTPS pool URL without credentials or a query.")
+    session_id = str(uuid4())
+
+    async with AzureCliCredential() as credential, httpx.AsyncClient(timeout=70, follow_redirects=False) as http_client:
+
+        @tool
+        async def execute_python(
+            code: Annotated[str, "Python code to execute in the remote session."],
+        ) -> dict[str, Any]:
+            """Run Python in the conversation's cloud sandbox and return its output and execution status."""
+            token = await credential.get_token("https://dynamicsessions.io/.default")
+            response = await http_client.post(
+                f"{pool_endpoint}/executions",
+                params={"api-version": "2025-10-02-preview", "identifier": session_id},
+                headers={"Authorization": f"Bearer {token.token}"},
+                json={
+                    "codeInputType": "Inline",
+                    "executionType": "Synchronous",
+                    "code": code,
+                    "timeoutInSeconds": 30,
+                    "outputStreamsMaxLength": 4096,
+                },
+            )
+            response.raise_for_status()
+            return response.json()
+
+        async with Agent(
+            client=FoundryChatClient(
+                project_endpoint=os.environ["FOUNDRY_PROJECT_ENDPOINT"],
+                model=os.environ["FOUNDRY_MODEL"],
+                credential=credential,
+            ),
+            instructions="Use execute_python for calculations. Check its status and errors before reporting a result.",
+            tools=[execute_python],
+        ) as agent:
+            result = await agent.run("Use Python to calculate the total of 19.95, 42.50, and 8.75, then add 8% tax.")
+            print(result.text)
+
+
+if __name__ == "__main__":
+    asyncio.run(main())