setup-foundry: Скилл для развертывания и настройки

--- name: setup-foundry description: | Provision a Microsoft Foundry resource, create a project, deploy a model (gpt-4.1-mini), and install the required Python SDK packages. This is a one-time setup skill. Use when the user says: "setup foundry", "provision foundry", "create foundry resource", "setup azure ai foundry", "install foundry sdk" metadata: verb: setup --- # Microsoft Foundry Setup Provision a Foundry resource, project, and model deployment via Azure CLI, then install the Python SDK. ## Step 1 -- Verify Azure CLI ```bash az version --output table ``` If the CLI is not installed, tell the user to install it from https://learn.microsoft.com/cli/azure/install-azure-cli and come back. ## Step 2 -- Authenticate ```bash az account show --output table ``` If not logged in, run: ```bash az login ``` ## Step 3 -- Choose Configuration Ask the user for the following values (or offer sensible defaults): | Setting | Default | |---------|---------| | Resource group name | `foundry-agents-rg` | | Foundry resource name | `foundry-<username>` (must be globally unique) | | Project name | `default` | | Location | `eastus` | | Model to deploy | `gpt-4.1-mini` | | Deployment name | `gpt-4.1-mini` | ## Step 4 -- Create the Resource Group ```bash az group create --name <RESOURCE_GROUP> --location <LOCATION> ``` ## Step 5 -- Create the Foundry Resource ```bash az cognitiveservices account create --name <FOUNDRY_RESOURCE_NAME> --resource-group <RESOURCE_GROUP> --kind AIServices --sku S0 --location <LOCATION> --allow-project-management --yes ``` ## Step 6 -- Set a Custom Subdomain The custom subdomain is required for SDK access. It must be globally unique. ```bash az cognitiveservices account update --name <FOUNDRY_RESOURCE_NAME> --resource-group <RESOURCE_GROUP> --custom-domain <FOUNDRY_RESOURCE_NAME> ``` ## Step 7 -- Create the Project ```bash az cognitiveservices account project create --name <FOUNDRY_RESOURCE_NAME> --resource-group <RESOURCE_GROUP> --project-name <PROJECT_NAME> --location <LOCATION> ``` ## Step 8 -- Deploy a Model ```bash az cognitiveservices account deployment create --name <FOUNDRY_RESOURCE_NAME> --resource-group <RESOURCE_GROUP> --deployment-name <DEPLOYMENT_NAME> --model-name <MODEL_NAME> --model-version "2025-04-14" --model-format OpenAI --sku-capacity 10 --sku-name GlobalStandard ``` If the model version or SKU is not available in the selected region, try `Standard` SKU or a different model version. You can list available models with: ```bash az cognitiveservices account list-models --name <FOUNDRY_RESOURCE_NAME> --resource-group <RESOURCE_GROUP> --output table ``` ## Step 9 -- Install the Python SDK ```bash pip install azure-ai-projects --pre openai azure-identity python-dotenv ``` ## Step 10 -- Verify the Setup Build the project endpoint from the resource and project names: ``` https://<FOUNDRY_RESOURCE_NAME>.services.ai.azure.com/api/projects/<PROJECT_NAME> ``` Run a quick verification: ```bash python3 -c " import os from azure.identity import DefaultAzureCredential from azure.ai.projects import AIProjectClient endpoint = 'https://<FOUNDRY_RESOURCE_NAME>.services.ai.azure.com/api/projects/<PROJECT_NAME>' client = AIProjectClient(endpoint=endpoint, credential=DefaultAzureCredential()) openai_client = client.get_openai_client() resp = openai_client.responses.create( model='<DEPLOYMENT_NAME>', input='Say hello in one sentence.', ) print('Model response:', resp.output_text) print('Setup verified successfully.') " ``` Replace the placeholders with the actual values chosen earlier. ## Step 11 -- Save Environment Variables Tell the user to note down or export these values for future use: ```bash export FOUNDRY_PROJECT_ENDPOINT="https://<FOUNDRY_RESOURCE_NAME>.services.ai.azure.com/api/projects/<PROJECT_NAME>" export FOUNDRY_MODEL_DEPLOYMENT_NAME="<DEPLOYMENT_NAME>" ``` ## Summary Tell the user: - The Foundry resource, project, and model deployment are ready - The project endpoint URL (they will need it for the other skills) - The model deployment name - The Python SDK packages are installed - They can now use the Foundry Code Interpreter and Foundry Agent Chat skills