Azure OpenAI Capacity Finder
Find available Azure OpenAI model capacity across regions and projects before choosing a deployment location.
The skill explicitly defines read-only analysis, avoids deployment, and requires project confirmation before handoff. The scripts appear to read subscription, capacity, project, and quota metadata, with no evidence of credential theft, covert exfiltration, or destructive defaults. Points are deducted because least-privilege Azure RBAC requirements, sensitive-data handling, data-flow disclosure, and rollback boundaries are not specified, and execution depends on the user's existing az login state.
The workflow, parameters, error table, and Bash scripts are broadly consistent; set -euo pipefail and several authentication, empty-capacity, and missing-project messages are present. Points are deducted because this is static-only review with no execution, the PowerShell implementations are not supplied, abnormal API/jq/Python/region/quota responses receive limited handling, and service responses are embedded into Python source through triple-quoted strings.
Trigger conditions and non-fit cases are relatively clear for capacity discovery, quota analysis, and region comparison, and the skill explicitly excludes deployment and quota requests. Points are deducted because the full discovery script hardcodes GlobalStandard, project discovery is limited to AIServices resources, the default-latest-version rule is coarse, and there is no Chinese-language guidance or mainland-China Azure cloud reachability disclosure.
The document has clear sections for quick reference, triggers, scripts, workflow, errors, related skills, MIT licensing, author, and version 1.0.0. Points are deducted because no changelog, explicit maintenance owner, or update path is provided; the referenced PowerShell scripts and parent-router rules are absent from the supplied evidence; and parameter constraints and examples remain incomplete.
The scripts are intended to produce a ranked region/project table with capacity, quota, and project information, followed by deployment or quota handoff, so the core task appears useful on the happy path. Points are deducted because static review cannot verify API compatibility or correctness, the complete discovery path is limited to GlobalStandard, and results still require manual validation of SKU, quota, and project selection.
The supplied material includes a substantial Bash implementation, usage examples, error behavior, and repository-level CI/testing context, providing some auditability. Points are deducted because no skill-specific acceptance scenarios or third-party execution evidence are supplied, and the cited repository tests cover the generic Copilot harness rather than independently reproducing this skill's key paths.
- The full discovery path is hardcoded to GlobalStandard and should not be treated as coverage for other SKUs.
- Before use, verify the Azure CLI identity, subscription scope, cloud environment, and jq/Python dependencies.
- Platform capacity, subscription quota, and existing projects are separate constraints; manually review the output before deployment handoff.
- The supplied evidence lacks the PowerShell script contents, skill-specific acceptance tests, and mainland-China reachability guidance.
What does this skill do, and when should you use it?
This read-only skill is part of Microsoft's Agent Skills collection and focuses on pre-deployment capacity discovery for Azure OpenAI models. It checks accessible regions and projects, analyzes model capacity alongside subscription quota, and ranks candidate locations. It does not deploy models or submit quota-increase requests. After discovery, it hands the user off to the preset or customize skill for deployment.
It first validates the signed-in Azure account with Azure CLI and can query available model versions when the requested version is unknown. Bash or PowerShell discovery scripts then call the relevant REST APIs, inspect capacity across accessible regions, and cross-reference the results with existing projects. The output is ranked by whether the target is met, project count, and available capacity, then annotated with subscription quota availability. It presents region, project, capacity, target status, and quota information before offering deployment or another search.
- An Azure OpenAI user needs to know where gpt-4o can be deployed.
- A team needs a region and project with at least 10K TPM for a specified model.
- A user received a quota error and needs to compare alternative regions.
- A user wants the best region and project for deploying a particular model.
What are this skill's strengths and limitations?
- Searches across accessible regions and projects instead of checking only one location.
- Separates platform capacity from subscription quota and flags candidates with zero available quota.
- Provides both Bash and PowerShell scripts with an explicit ranking method.
- Remains read-only and does not deploy the model automatically.
- Requires Azure CLI authentication and suitable Azure access.
- Does not perform deployment or submit quota-increase requests.
- The supplied material gives no dedicated test-coverage or success-rate evidence for this skill.
- When no project is found, the user must follow a separate project-creation path or use the Azure Portal.
How do you install this skill?
Install the collection with npx skills add microsoft/skills, then select the required skill in the wizard. Skills are installed into the selected agent's directory, such as .github/skills/ for GitHub Copilot. In Copilot CLI, the supported plugin commands are /plugin marketplace add microsoft/skills followed by /plugin install azure-skills@skills. The source does not document a separate installation procedure for this capacity sub-skill.
How do you use this skill?
Run az login, then ask something such as “Find a region where I can deploy gpt-4o with 10K TPM capacity.” The skill validates the Azure account; if necessary, it runs ./scripts/query_capacity.sh <model-name> or the PowerShell equivalent, followed by ./scripts/discover_and_rank.sh <model-name> <version> <min-capacity> or its PowerShell equivalent. After discovery, confirm the target project and choose quick deployment, custom deployment, another model or capacity search, or cancellation. The source does not specify required Azure roles or the exact REST response schema.