agent-framework/dotnet/samples/04-hosting/FoundryHostedAgents/responses/Hosted-AgentSkills

What this sample demonstrates

An Agent Framework agent that loads its behavioral guidelines from Foundry Skills at startup, hosted using the Responses protocol. Skills are authored once as SKILL.md files, uploaded to your Foundry project through the Skills REST API, and downloaded by the agent on boot so updates ship without code changes.

How It Works

Authoring skills

Each skill is a Markdown file with a YAML front matter block. This sample ships two source skills under skills/:

Skill	Purpose
support-style	Voice, formatting, and signature rules for Contoso Outdoors support replies.
escalation-policy	When and how to escalate a customer ticket.

Each SKILL.md includes a unique *-CANARY-* token that the model is asked to echo, so you can prove the skill was loaded from Foundry (not hallucinated) by checking the response.

The name and description values in the YAML front matter must be unquoted — quoting them causes the Skills REST API to return HTTP 500 on import.

Uploading skills

The sample includes a convenience provisioning step that checks whether each skill exists in Foundry and uploads it if not, gated behind the PROVISION_SAMPLE_SKILLS=true env var. In production, skill provisioning is an external concern — it is NOT the hosted agent's responsibility. A real deployment pipeline would provision skills separately (e.g., via a CI/CD step, a CLI script, or a management portal).

The provisioning uses ProjectAgentSkills.CreateSkillFromPackageAsync(directoryPath) from the Azure.AI.Projects.Agents SDK. The method packages the SKILL.md file as a ZIP and uploads it to Foundry.

Downloading skills at agent startup

Program.cs reads the comma-separated SKILL_NAMES env var and for each skill name downloads the ZIP archive from Foundry via ProjectAgentSkills.DownloadSkillAsync(name), then unpacks it into a separate runtime directory at downloaded_skills/<name>/ (kept distinct from the static skills/ source folder).

An AgentSkillsProvider is then built over downloaded_skills/ and attached to the agent as a context provider. The provider follows the Agent Skills progressive-disclosure pattern:

1. Advertise — skill names and descriptions are injected into the system prompt at session start (~100 tokens per skill).
2. Load — the model calls the load_skill tool when it decides a skill is relevant to the user's turn, and the full SKILL.md body is returned.

This means the model only pays the token cost for a skill's full body when it actually needs it, and updating a skill in Foundry + restarting the agent is enough to pick up the change — no code redeploy required.

Note: This sample supports instruction-only and resource-based skills. If your downloaded skills contain scripts, add a script runner when constructing the AgentSkillsProvider.

Agent Hosting

The agent is hosted using the Agent Framework with the Responses API hosting layer (AddFoundryResponses / MapFoundryResponses).

Prerequisites

• An Azure AI Foundry project with a deployed model (e.g., gpt-4o)
• Azure CLI logged in (az login)

Required RBAC

Your identity (or the Managed Identity running the container in production) needs Azure AI User on the Foundry project scope. This single role covers both authoring skills and downloading them.

Running the Agent Host

Set the required environment variables and run the sample with dotnet run:

export AZURE_AI_PROJECT_ENDPOINT="https://<account>.services.ai.azure.com/api/projects/<project>"
export AZURE_AI_MODEL_DEPLOYMENT_NAME="gpt-4o"
export SKILL_NAMES="support-style,escalation-policy"
export PROVISION_SAMPLE_SKILLS="true"   # First run only — provisions skills to Foundry

Or in PowerShell:

$env:SKILL_NAMES="support-style,escalation-policy"
$env:PROVISION_SAMPLE_SKILLS="true"     # First run only — provisions skills to Foundry

You can also place these in a .env file next to Program.cs — see .env.example.

On startup you should see:

Skill 'support-style' already exists in Foundry.
Skill 'escalation-policy' already exists in Foundry.
Downloading skill 'support-style' from Foundry...
Downloading skill 'escalation-policy' from Foundry...

The downloaded SKILL.md files land under downloaded_skills/<name>/SKILL.md next to the published output. This directory is recreated from scratch on every run, so deleting it manually is never necessary.

Interacting with the agent

Send a POST request to the server with a JSON body containing an "input" field to interact with the agent. For example:

curl -X POST http://localhost:8088/responses -H "Content-Type: application/json" -d '{"input": "Hi, I am Alex. I just want to confirm I can return my tent within 30 days."}'
curl -X POST http://localhost:8088/responses -H "Content-Type: application/json" -d '{"input": "I want a $750 refund on Order #A-1042 right now or I am calling my lawyer."}'

Prompt mentions	Skill that should drive the response
Routine return / shipping / care question	Model loads support-style (canary STYLE-CANARY-3318) — no escalation.
Injury, legal threat, press, or refund > $500	Model loads escalation-policy (canary ESC-CANARY-7742) and support-style.

Because skills are loaded on demand, the canary token in a response also proves the model actually invoked load_skill for the matching skill (not just saw its name in the advertised list).

Deploying the Agent to Foundry

When deploying to Foundry, make sure SKILL_NAMES is set in your azd environment so it gets injected into the hosted container per agent.manifest.yaml:

azd env set SKILL_NAMES "support-style,escalation-policy"

The deployed agent's Managed Identity needs Azure AI User on the Foundry project to download skills at startup.

The skills/ source folder is not deployed to Foundry — only the downloaded skills are used at runtime. The provisioning step must have been run against the same Foundry project before the agent can download the skills.