# Manage Microsoft Foundry Agents

Foundry agent projects require coordinated setup across azd, Azure resources, models, tools, and evaluations. This skill guides Claude, Codex, and Claude Code through those workflows with project-specific checks and handoffs.

## Install

```bash
npx skillstore add microsoft/microsoft-foundry
```

## Metadata

- Status: approved
- Slug: microsoft-microsoft-foundry
- Version: 0.0.0-placeholder
- Author version: 0.0.0-placeholder
- Skillstore revision: r1
- Version status: valid
- Tree hash: 7fe2ce83bf57ad86cedba75be323f25f48d26419a781ce2f14fc029ef6839a41
- Author: microsoft
- GitHub username: microsoft
- License: MIT
- Repository: https://github.com/microsoft/github-copilot-for-azure/tree/main/plugin/skills/microsoft-foundry/
- Ref: 1200d2f84325e4a40a4c6cc230f3864179fd6940
- Supported tools: Claude, Codex, Claude Code
- Audit status: complete
- Agent install advisory: confirmation\_required
- Manual install advisory: allowed
- Artifact signature: available
- Audit attestation: unavailable
- Human verification: not\_verified
- Risk factors: network, env\_access, scripts, external\_commands, filesystem
- Quality score: 76
- Public page: https://skillstore.pages.dev/skills/microsoft-microsoft-foundry
- Manifest: https://skillstore.pages.dev/api/skills/microsoft-microsoft-foundry/manifest

## Capabilities

- Route Foundry requests to focused sub-skills for create, deploy, invoke, observe, trace, troubleshoot, and routine workflows.
- Guide hosted agent scaffolding and deployment with Azure Developer CLI and Foundry project context.
- Support evaluation setup, batch runs, continuous evaluation, prompt optimization, trace analysis, and dataset curation.
- Assist model deployment planning across presets, custom settings, capacity, quota, and region choices.
- Provide fine-tuning guidance for SFT, DPO, RFT, graders, training data, checkpoints, and deployment.
- Handle RBAC, project creation, resource creation, private networking, and connection setup guidance.

## Use Cases

- Create a Hosted Agent: Scaffold a hosted Foundry agent, connect it to a project, deploy it, and run a smoke test.
- Evaluate Agent Quality: Build evaluation datasets, run batch evaluations, inspect traces, and optimize prompts or instructions.
- Prepare Production Infrastructure: Plan resources, model capacity, quota, RBAC, networking, and deployment settings for Foundry workloads.

## Prompt Templates

### Start a Foundry Project

```
Help me create a new Azure AI Foundry project for a hosted agent. Ask for any missing subscription, region, and project details first.
```

### Deploy a Hosted Agent

```
Review my current Foundry agent project, verify the azd environment, and guide me through a safe deployment and smoke test.
```

### Improve Agent Quality

```
Set up an evaluation workflow for my Foundry agent, suggest datasets and evaluators, then help analyze failures and optimize instructions.
```

### Plan Production Rollout

```
Create a production readiness plan for a Foundry agent, including RBAC, quota, model capacity, private networking, evaluation, and rollback checks.
```

## Limitations

- It does not replace Azure portal permissions, billing, quota approvals, or organization policy decisions.
- Many workflows require Azure CLI, azd, extensions, local project files, and authenticated user context.
- It can propose and run commands only when the host tool permits command execution.
- It should not be used for general Azure deployment outside Foundry-focused workflows.

## Best Practices

- Confirm the target subscription, resource group, project endpoint, agent root, and environment before write operations.
- Prefer managed identity and project connections over raw API keys when a workflow supports both.
- Keep generated evaluation artifacts under the selected .foundry folder and review them before deployment.

## Anti Patterns

- Do not run deployment, quota, RBAC, or cleanup commands against an inferred subscription.
- Do not paste secrets into prompts when managed identity, environment variables, or project connections can be used.
- Do not reuse stale azd environment values across development, staging, and production projects.

## Security Audit

- Audited at: 2026-07-20T03:30:58.619\+00:00
- Summary: All 400 representative static findings were adjudicated as false positives in context. They describe documented Azure Foundry administration, local project configuration, or guarded utility behavior; no prompt injection, credential exfiltration, or malicious intent was found. The skill can still execute cloud administration workflows, so users should review commands before applying changes. Static review was capped at 400/1141 representative findings; omitted static matches are unconfirmed, so automatic publishing stays disabled until manual review.

## Stats

- Views: 208
- Downloads: 8
- Favorites: 0
- Popularity score: 0