Azure OpenAI Deployment Assistant
Discover capacity and deploy Azure OpenAI models through intent-based routing.
The evidence requires confirmation of the project, region, and full configuration before deployment, validates model/SKU/quota data, and clearly labels the operation as an out-of-band Azure change. Points are deducted because the required Azure permissions are substantial and deployments may incur ongoing cost; the Anthropic flow reads tenant country and organization data and sends them in the deployment request without sufficient data-flow explanation; rollback is only documented as a delete command, not as a confirmed recovery procedure.
The documentation provides prerequisite checks, input constraints, cross-region fallback, polling, and error handling, making the happy path plausible. Static calibration caps this below 10; points are deducted for inconsistencies such as AIServices versus AIProject project filters, scripts centered on GlobalStandard despite broader documented SKU support, and fragile embedded Python/JSON interpolation for unusual model names. Execution was not reproduced statically.
Intent keywords, three operating modes, negative triggers, and the azd-managed non-fit boundary are clearly described. Points are deducted because Chinese-language interaction, semantic edge boundaries, and CLI/MCP/Portal selection conditions are unspecified; mainland-China network reachability is not established, while core operation depends on Azure management endpoints and an authenticated CLI environment.
The material provides layered entry documentation, quick references, sub-skills, scripts, examples, guides, workflow details, and troubleshooting, with MIT licensing, authorship, and version metadata. Points are deducted because versions differ between the router and customize sub-skill (1.0.0 versus 1.0.1), maintenance ownership and a changelog are not clear in the selected skill, the workflow claims 14 phases but lists 13, and some documented capabilities exceed the evident script scope.
The skill addresses the full stated deployment task from project selection and model/SKU/capacity validation through deployment monitoring, and its examples and test prompts describe directly usable flows. Static calibration limits effectiveness to 7; points are deducted because representative execution is not independently verifiable, project discovery and SKU/region coverage have implementation concerns, and users may still need manual correction.
Committed scripts, detailed commands, test prompts, and reported automated results provide some auditability. Static calibration limits verifiability to 5; points are deducted because the test results are author-reported, no independently verifiable key-path test suite or third-party reproduction evidence is supplied for this skill, and nothing was executed in this review.
- Deployment changes external Azure resources and may incur cost; confirm the project, region, SKU, capacity, and final configuration before execution.
- Recheck the AIServices versus AIProject filters and the scripts' apparent GlobalStandard-only handling, which may omit or misselect targets.
- Treat the tenant country code and organization name collected by the Anthropic flow as potentially sensitive and verify compliance and purpose before sending them.
- No evidence establishes Chinese-language support or mainland-China network reachability; verify Azure CLI authentication and management API access in the target environment.
What it does & when to use it
This skill is a unified entry point for Azure AI Foundry model deployment workflows. It supports preset deployments, customized deployments, and capacity discovery across regions and projects. It interprets the request, resolves the target project, and validates model SKU support and subscription quota before deployment. It is intended for Foundry projects outside the azd lifecycle and for ad-hoc deployments; it does not list or delete deployments, create agents, or create projects.
Interprets requested models, regions, capacity, SKUs, versions, and RAI policies; resolves a project from the PROJECT_RESOURCE_ID environment variable, the prompt, or project discovery; routes requests to preset, customize, or capacity-discovery modes; uses Azure CLI, MCP, or the portal for deployment workflows; runs az cognitiveservices model list to verify supported SKUs and az cognitiveservices usage list to calculate available quota; and presents the target project for confirmation before deployment.
- An Azure developer wants to deploy a GPT model quickly without custom settings.
- A Foundry administrator needs to choose a model version, SKU, capacity, or RAI policy.
- An Azure user needs to find regions or projects with enough quota for a model deployment.
- A team wants to discover a suitable region first and then deploy with either defaults or custom settings.
- An engineer needs an ad-hoc deployment outside an azd-managed project lifecycle.
Pros & cons
- Combines preset deployment, customization, and cross-region capacity discovery behind one entry point.
- Dynamically validates model SKU support and available subscription quota before presenting deployment options.
- Requires explicit target-project confirmation before deployment.
- Clearly defines its boundary with azd-managed Foundry projects.
- Requires Azure CLI authentication, an Azure subscription, and deployment permissions.
- The workflow can use Azure CLI, MCP, or the portal, but the supplied source does not document specific MCP tool names or complete failure handling.
- Only the main router skill is provided; detailed sub-skill procedures are not included in the supplied material.
- It does not list existing deployments, delete deployments, create agents, or create projects.
How to install
Install the repository's skill collection with npx skills add microsoft/skills, then select the needed skills in the wizard. Skills are installed into the selected agent directory, such as .github/skills/ for GitHub Copilot. The source does not document a separate installation command for this nested skill or provide a standalone package.
How to use
Install and authenticate Azure CLI, and have an Azure subscription, deployment permissions, and an Azure AI Foundry project resource ID available. Use prompts such as “deploy gpt-4o”, “deploy gpt-4o with custom settings”, or “find where I can deploy this model” to trigger preset, customized, or capacity-discovery routing. Before deployment, the skill must confirm the target project. For azd-managed projects, declare deployments in azure.yaml under services.ai-project.deployments[] instead of using this skill.