Search skills/

Microsoft Foundry (Azure Skills) skill

Microsoft's skill for building, deploying, evaluating, fine-tuning and troubleshooting Foundry agents and models, routed through a dozen sub-skills.

in microsoft/azure-skills Planning, TDD & debuggingCloud & DevOps
Our verdict Recommended with caveats

Deepest Foundry skill available and from Microsoft, but it assumes azd, the Foundry MCP and an Azure subscription, and its mandatory-step style makes sessions long.

4/5 Depth expertise shipped ! Security Broad access 0d Maintenance Updated today 582K Adoption installs OK License MIT, commercial OK

What the microsoft-foundry skill does

This is the largest skill in Microsoft's Azure plugin: a 31 KB router SKILL.md plus about 120 files covering Microsoft Foundry end to end. With it loaded, the agent first runs a dependency check script that installs the azd microsoft.foundry extension if it is missing, then reads the matching sub-skill before touching any workflow. The sub-skills cover creating hosted, prompt, managed-harness and voice agents, deploying and invoking them, scheduled routines, CI/CD, trace queries through App Insights, hosted-agent log troubleshooting, batch and continuous evaluation, and an Agent Optimizer that rewrites prompts from eval results. A separate finetuning folder handles SFT, DPO and RFT jobs with 17 Python scripts (dataset validation and conversion, grader calibration, job submission, training monitoring, model deployment) and reference notes on reward hacking and training curves. Other folders cover model deployment, quota, RBAC and resource creation. The instructions are written as mandatory step lists with gates and expect the Foundry MCP tools and the azd CLI to be available.

Good for

  • Teams building agents on Microsoft Foundry who want azd-based scaffold, deploy and eval flows
  • Fine-tuning projects on Azure that need dataset validation and grader tooling
  • Debugging failed Foundry deployments, quota or RBAC problems

Not for

  • General Azure app deployment (use azure-prepare and azure-deploy)
  • Projects not using Foundry or Azure

Depth 4/5

31 KB router plus roughly 120 files: 87 reference and workflow docs and 27 scripts including a Python fine-tuning toolkit and validators, though no evals for the skill itself.

Dependencies and costs

  • Azure account or subscription
  • Azure CLI signed in with az login
  • Azure MCP Server (installed with the plugin; needs Node.js 18+ for npx)
  • Azure Developer CLI (azd) with the microsoft.foundry extension (installed automatically by the dependency script)
  • Python for the fine-tuning scripts
  • A Microsoft Foundry project; training and model deployments consume Azure quota and credits
  • Telemetry: The skill files contain no telemetry code. The Azure MCP Server that the plugin wires in has its own telemetry; the repo README says to disable it with AZURE_MCP_COLLECT_TELEMETRY=false.

Security review

Broad access

This skill asks for more access than most, or sends data to another service. That is usually a design choice rather than bad intent, but you should know before you install it.

  • Pre-approves broad shell commandsFrontmatter lists unrestricted 'allowed-tools: Read, Write, Bash, AskUserQuestion' (private-network.md line 5 adds microsoft_docs_fetch); these are sub-skill docs, not the root SKILL.md, so a host may ignore them..github/plugins/azure-skills/skills/microsoft-foundry/project/create/create-foundry-project.md:7

A large Foundry skill (agents, model deployment, evaluation, fine-tuning, networking) with scripts that call Azure Resource Manager and your Foundry or Azure OpenAI endpoint using your Azure login or API key, upload training data you point at, and delete or cancel jobs on request. Two sub-skill docs list unrestricted Bash in allowed-tools, and the grader calibration script runs a Python file you supply.

Worth knowing

  • Relies on a CLI or package we could not reviewNeeds azd, az and Python SDKs (openai, azure-identity, requests, websockets) that are not part of the skill..github/plugins/azure-skills/skills/microsoft-foundry/scripts/check-and-setup-dependencies.sh:23
  • Installs packages or toolsRuns 'azd extension install microsoft.foundry' when it is missing..github/plugins/azure-skills/skills/microsoft-foundry/scripts/check-and-setup-dependencies.sh:23
  • Makes network requestsCalls Azure Resource Manager (management.azure.com) with an az-issued token to create, list and delete model deployments; other scripts upload training files and submit jobs to your Foundry or Azure OpenAI endpoint; create-foundry-project.md line 78 runs 'azd init -t' on the Azure-Samples/azd-ai-starter-basic template..github/plugins/azure-skills/skills/microsoft-foundry/finetuning/scripts/deploy_model.py:120
  • Needs an API key for its own serviceAccepts AZURE_OPENAI_API_KEY or --api-key and uses it only against the Azure OpenAI or Foundry endpoint you supply; otherwise uses DefaultAzureCredential..github/plugins/azure-skills/skills/microsoft-foundry/finetuning/scripts/common.py:106

What we checked

The skill folder (221 files) on Oct 29, 2026. Not a full review: 101 of 221 files were not downloaded (large repository; we reviewed the skill folders); we read 11 of 27 code files; it relies on code outside the repository: Azure CLI (az) and the az extensions it installs on demand; not part of the files we have; Azure Developer CLI (azd) and its azure.ai.agents, azure.ai.projects and microsoft.foundry extensions; Azure MCP Server, started by the plugin through npx; its configuration sits outside this skill folder.

NVIDIA SkillSpector static scan: 61 pattern hits (Excessive Agency 15, analysis-evasion 13, Data Exfiltration 9, Dangerous Code Execution 5, Data Flow 5, Privilege Escalation 4). Two CRITICAL hits are by design: calibrate_grader.py exec()s a grader Python file you pass in (its docstring warns to use only trusted files) and deploy_model.py sends an az-issued ARM token to management.azure.com. The prompt-injection hit on 'Send messages to' is the invoke sub-skill name, not an instruction to exfiltrate.

A review, not a certification. Skills change after we check them — read the files before installing.

License

MIT (declared in the SKILL.md frontmatter); the repo is a mirror synced from microsoft/GitHub-Copilot-for-Azure and has no top-level LICENSE file in our copy. Commercial use allowed under MIT.

Can it run unattended?

Not really. Workflows stop at gates for user approval and use real cloud resources, so they are meant to be supervised, although the fine-tuning monitor scripts can poll long-running jobs.

Alternatives

Google agents-cli

google/agents-cli

Google's CLI and seven skills that teach a coding agent to scaffold, build, evaluate, deploy and observe ADK agents on Google Cloud.

With caveats Check first Depth 4/5

Hugging Face Skills

huggingface/skills

Official Hugging Face skills led by hf-cli, a Hub CLI reference generated from your local install, plus AWS/SageMaker deployment and operations skills.

Recommended Check first Depth 3/5

Google Agent Skills

google/skills

Google's official collection of about 180 skills for Google Cloud, Gemini APIs, Analytics, Data Manager, sign-in and other Google products.

Recommended Broad access Depth 3/5

FAQ

What is the Microsoft Foundry skill?

It is one of 17 skills in Microsoft's azure-skills plugin. It teaches an agent how to create, deploy, evaluate, optimize, fine-tune and troubleshoot agents and models in Microsoft Foundry using the azd CLI and Foundry MCP tools.

How do I install the Microsoft Foundry skill?

In Claude Code run /plugin install azure@claude-plugins-official; in Codex add the marketplace with codex plugin marketplace add microsoft/azure-skills and install the azure plugin. You need Node.js 18+, the Azure CLI signed in with az login, and azd for deployment workflows.

Does the Foundry skill install anything on my machine?

Its first mandatory step runs a dependency script that checks for azd and installs the microsoft.foundry azd extension if it is not present. It does not reinstall dependencies that already exist.

Foundry skill vs azure-deploy: which one do I use?

Use the Foundry skill for anything about Foundry agents, models, evals and fine-tuning. Use azure-prepare, azure-validate and azure-deploy for general Azure apps; the Foundry skill itself says to route general Azure deployment to those skills.

Sources: GitHub repository; install counts from skills.sh via Best Skills (CC BY 4.0). Review based on the published SKILL.md.