Microsoft Foundry

Microsoft Foundry Skill

This skill helps developers work with Microsoft Foundry resources, covering model discovery and deployment, complete dev lifecycle of AI agent, evaluation workflows, and troubleshooting.

Pre-Execution Requirements

Follow each applicable subsection below before starting its corresponding action or workflow.

Dependency Check and Setup

MANDATORY: As the first step after this skill loads, run the dependency check and setup script below from this skill's root and wait for it to finish before continuing. The script checks first and installs only missing dependencies; it does not reinstall dependencies that are already available.

You MUST complete this check before reading or entering any sub-skill, workflow, or workflow-specific reference.

./scripts/check-and-setup-dependencies.sh     # macOS / Linux
./scripts/check-and-setup-dependencies.ps1    # Windows (pwsh)

Strictly follow the script output for subsequent actions.

Workflow Guidance

MANDATORY: Before executing ANY workflow-specific steps, you MUST read the corresponding sub-skill document. Do not call workflow-specific MCP tools for a workflow without reading its skill document. This applies even if you already know the MCP tool parameters — the skill document contains required workflow steps, pre-checks, and validation logic that must be followed. This rule applies on every new user message that triggers a different workflow, even if the skill is already loaded.

Foundry MCP

MANDATORY: Before using Foundry MCP operations, call the Azure MCP foundry tool and inspect the available Foundry MCP tools and related parameters. Treat this as the discovery/help step for MCP-based workflows.

azd

MANDATORY: Before executing ANY azd command, you MUST read azd-guidance and strictly follow the shared rules defined in it, especially the AZURE_DEV_USER_AGENT setting rules.

Sub-Skills

This skill includes specialized sub-skills for specific workflows. When a sub-skill matches the task, strictly follow its workflow:

💡 Tip: For a complete onboarding flow: project/create (public) or private-network (VNet isolation) → models/deploy-model → agent workflows (create → deploy → invoke).

💡 Fine-Tuning: Use finetuning for all model customization — SFT distillation, DPO preference optimization, and RFT with graders. Includes quickstart, grader calibration, and training curve analysis.

💡 Model Deployment: Use models/deploy-model for all deployment scenarios — it intelligently routes between quick preset deployment, customized deployment with full control, and capacity discovery across regions.

💡 Prompt Optimization: For requests like "optimize my prompt" or "improve my agent instructions," load observe and use the prompt_optimize MCP tool through that eval-driven workflow.

Infrastructure Lifecycle

Match user intent to the correct infrastructure workflow.

Agent Development Lifecycle

Match user intent to the correct agent workflow. Read each sub-skill in order before executing.

Agent: .foundry Workspace Standard

Every agent source folder can keep Foundry-specific cache and overlay state under .foundry/:

<agent-root>/
  .foundry/
    agent-metadata.yaml
    agent-metadata.prod.yaml
    suites/
    datasets/
    evaluators/
    results/

• In azd projects, derive deployment context (project endpoint, agent name/version, ACR, App Insights) from azure.yaml plus azd env get-values; do not duplicate those values in metadata when azd already provides them.

• agent-metadata.yaml is the preferred local/dev overlay for non-azd values, remote Foundry suite references, local cache paths, result summaries, and explicit overrides. Optional sidecar files such as agent-metadata.prod.yaml can hold a single prod or CI-targeted overlay without mixing multiple environments in one file.

• suites/, datasets/, and evaluators/ are local cache folders. Reuse them when they are current, and ask before refreshing or overwriting them.

• See Agent Metadata Contract for the canonical schema and workflow rules.

Agent: Setup References

• Standard Agent Setup — advanced setup for production workloads that need data-residency control (bring-your-own Cosmos DB / Storage / AI Search via a Foundry capability host). The default azd ai agent flow uses Basic Agent Setup and does not provision capabilityHosts/agents — do not flag its absence as a bug. For default post-provision state, see the "Expected env-var fingerprint" section in foundry-agent/create/create-hosted.md.

Agent: Common Project Context Resolution

Agent skills should run this step only when they need configuration values they don't already have. If a value (for example, agent root, environment, project endpoint, or agent name) is already known from the user's message or a previous skill in the same session, skip resolution for that value.

Step 1: Discover Agent Roots and azd Context

First check whether the workspace has azure.yaml with services using host: azure.ai.agent.

• One azd agent service -> classify it from kind, harness, modelType, and conversationEngine, then use that service's project folder or the azure.yaml directory as the agent root.

• Multiple azd agent services -> check whether a Voice wrapper references a kind: hosted service via conversationEngine.name:

  • Selecting the Hosted Voice wrapper implies its named kind: hosted target as an upstream dependency (uses: [<target>], conversationEngine.name: <target>); deploy and verification must report both services separately.

  • Selecting the kind: hosted target alone scopes changes/debugging to the target container and does not automatically modify the wrapper.

  • Otherwise -> require the user to choose the target service/folder.

• No azd agent service -> search the workspace for .foundry/ folders that contain agent-metadata.yaml or agent-metadata.<env>.yaml.

  • One match -> use that agent root.

  • Multiple matches -> require the user to choose the target agent folder.

  • No matches -> for insights, ask only for missing remote inputs without initializing local files; for create/deploy workflows, seed a new .foundry/ folder during setup; for other workflows, stop and ask the user which agent source folder to initialize.

After selecting an agent root, keep all local .foundry cache inspection, source inspection, evaluator suggestions, dataset suggestions, and prompt-optimization context inside that folder only. Do not scan sibling agent folders unless the user explicitly switches roots.

Step 2: Resolve Environment and Deployment Context

If azure.yaml is present, resolve the azd environment first:

1. Environment explicitly named by the user

2. AZURE_ENV_NAME from azd env get-values

3. azd default environment from .azure/config.json

4. Environment already selected earlier in the session

Run azd env get-values for the selected environment when project/deployment values are not already known. Prefer azd values for deployment context:

When azd supplies these values, use them as the source of truth and do not copy them into .foundry/agent-metadata*.yaml on metadata writes.

Step 3: Select Metadata Overlay and Resolve Environment

Inside the selected agent root, choose the metadata file in this order:

1. Metadata filename or path explicitly provided by the user or workflow

2. If an explicit environment is already known and .foundry/agent-metadata.<env>.yaml exists, use that file

3. .foundry/agent-metadata.yaml

4. If multiple metadata files remain and no rule above selects one, prompt the user to choose

Read the selected metadata file and resolve any remaining environment choice in this order:

1. Environment explicitly named by the user

2. If the selected metadata file defines exactly one environment, use it

3. Environment already selected earlier in the session

4. defaultEnvironment from metadata

If the selected metadata file still contains multiple environments and none of the rules above selects one, prompt the user to choose. Keep the selected agent root, metadata file, environment, and whether context came from azd or metadata visible in every workflow summary.

If the selected environment exposes older testSuites[] metadata but not evaluationSuites[], treat testSuites[] as the source for this session and normalize each entry in memory to the evaluationSuites[] shape before continuing. If the metadata is older still and only exposes legacy testCases[], normalize that list the same way. Preserve dataset and evaluator fields, keep any existing tags, and map legacy priority to tags.tier only when tags.tier is missing: P0 -> smoke, P1 -> regression, P2 -> coverage.

Step 4: Resolve eval.yaml Local Evaluation Intent

If eval.yaml exists in the selected agent root, parse it before generating new suites:

• agent.name -> target agent candidate; verify it matches the selected azd/metadata agent before using it.

• dataset.local_uri -> local seed dataset candidate; legacy dataset_file may be normalized in memory.

• dataset.name / dataset.version -> registered dataset candidate.

• validation_dataset -> optional validation dataset candidate.

• evaluators[] -> candidate Foundry evaluator names; verify with evaluator_catalog_get before treating them as remote evaluators.

• name -> local eval/suite candidate; verify remotely before persisting as suiteName.

• options.eval_model, options.optimization_model, options.max_candidates, options.optimization_config.model_search_space, options.pass_threshold, max_samples, trace_days, and generation_instruction -> setup defaults.

Treat eval.yaml as local evaluation intent, not proof that a Foundry suite exists. Persist synced suite/dataset/evaluator references to .foundry only after remote lookup or registration succeeds.

Step 5: Resolve Common Configuration

Layer sources in this order:

1. Explicit user input and values already selected in the session

2. azd environment values for deployment context

3. .foundry/agent-metadata*.yaml overlay values and remote suite/cache references

4. azure.yaml and eval.yaml local source configuration

5. User prompts for anything still missing

If azd and metadata both provide the same value and they differ, stop and ask which source is authoritative. If they match, use the azd value and avoid rewriting the duplicate on future metadata writes.

Step 6: Write Metadata Overlay (Create/Deploy/Observe Only)

On any metadata write (deploy, auto-setup, dataset refresh, or trace-to-dataset update), persist only non-derivable overlay/cache state in the selected metadata file:

• azd binding (azd.environmentName, azd.service) when useful for future resolution

• evaluationSuites[] with remote suite/dataset/evaluator references and local cache paths

• lastEval, result files, comparison summaries, or explicit non-azd overrides

Do not copy azd-owned deployment values into metadata when azd already provides them. If the selected file is a preferred single-environment file, rewrite only that one environment block. If the selected file is a legacy multi-environment file, rewrite only the selected environment block. Never copy or merge environments across sibling metadata files automatically. If the selected environment still uses older testSuites[] or legacy testCases[], rewrite it to evaluationSuites[] and remove migrated priority fields from the rewritten entries.

Step 7: Collect Missing Values

Use the ask_user or askQuestions tool only for values not resolved from the user's message, session context, metadata, or azd bootstrap. Common values skills may need:

• Agent root — Target azd service project folder or folder containing .foundry/agent-metadata*.yaml

• Metadata file — agent-metadata.yaml for local/dev, or an explicit sidecar such as agent-metadata.prod.yaml

• Environment — azd environment, dev, prod, or another environment key from metadata

• Project endpoint — Microsoft Foundry project endpoint URL

• Agent name — Name of the target agent

💡 Tip: If the user already provides the agent path, environment, project endpoint, or agent name, extract it directly — do not ask again.

Agent: Agent Types

Agent workflows distinguish four first-class agent types:

Classify an azure.yaml service by its definition in this exact order, not by host alone:

1. kind: voice -> Voice. For compatibility, silently classify an existing kind: prompt-voice service as Voice, then normalize it when editing its YAML. Classify the subtype:

   • conversationEngine.type: hosted_agent -> Hosted Voice wrapper (paired with a named kind: hosted target service).

   • modelType: managed or modelType: self_deployed -> Prompt Voice (Foundry infers native speech-to-speech vs cascaded STT -> LLM -> TTS from model.id).

   • Having both modelType/model and conversationEngine is invalid (stop and fix azure.yaml); having neither is incomplete (do not guess). Never use obsolete modelType: hosted_agent, targetAgent, or a nested config: block.

2. kind: hosted -> Hosted. When it declares protocols: [{ protocol: invocations_ws, version: 1.0.0 }] and metadata.voiceLiveCompatible: "true" alongside a kind: voice wrapper, it is a Hosted Voice target; otherwise treat a standalone invocations_ws service as a custom Hosted real-time/audio agent.

3. kind: prompt plus harness.type: microsoft_copilot_preview or legacy github_copilot_preview -> Managed Harness Agent.

4. kind: prompt without that harness -> ordinary Prompt Agent.

Use agent_get only when the type cannot be resolved from project context. Preserve the harness discriminator already authored on an existing Managed Harness Agent; do not migrate it during ordinary edits or redeploys. Do not convert an existing Agent between Prompt, Hosted, Managed Harness Agent, and Voice types; create a new Agent instead. Before creating any Voice Agent, determine the model strategy first per create-voice.

Tool Usage Conventions

• Use the ask_user or askQuestions tool whenever collecting information from the user

• Use the task or runSubagent tool to delegate long-running or independent sub-tasks (e.g., env var scanning, status polling, Dockerfile generation)

• Prefer azd for Hosted Agents, Managed Harness Agents, and Voice Agents; prefer Foundry MCP for ordinary Prompt Agents.

• Reference official Microsoft documentation URLs instead of embedding CLI command syntax

Azure Authentication

• Azure Authentication Best Practices

Additional Resources

• Foundry Hosted Agents

• Foundry Agent Runtime Components

Network Isolation Errors

Applies to any call against a Foundry project or its parent Foundry account — Foundry MCP tools, azd, az CLI, curl, REST, or SDK.

If an error matches Public access is disabled / PublicNetworkAccessDisabled / 403 Forbidden from a private endpoint / connection timeout / the project endpoint FQDN resolves to a public IP, this typically means the parent Foundry account has publicNetworkAccess=Disabled or Enabled from selected IP addresses, and the current shell is outside its VNet.

Only if the error is ambiguous, confirm against the Foundry account using a management-plane call (works from anywhere with reader access):

az cognitiveservices account show \
  --name <account> --resource-group <rg> \
  --query "properties.{publicNetworkAccess:publicNetworkAccess, networkAcls:networkAcls, privateEndpointConnections:privateEndpointConnections[].properties.privateLinkServiceConnectionState.status}"

publicNetworkAccess: "Disabled" — or "Enabled" together with non-empty networkAcls.ipRules / virtualNetworkRules — confirms isolation. If publicNetworkAccess: "Enabled" and networkAcls is empty, the failure is a caller-side network issue (e.g. Private DNS resolving the FQDN to a public IP from inside a VNet with a private endpoint), not an account-config issue.

If it's indeed a network isolation issue, supported connection options are documented in Choose a secure connection method to Foundry.

ℹ️ Foundry MCP tools cannot reach a VNet-isolated project even from inside the VNet.