AGENTIC MEMORY

Skills

Useful instructions for your AI. Search, read and use them through the same MCP connection.

← Back to Skills

microsoft-foundry

Build, deploy, evaluate, optimize, fine-tune, and manage Microsoft Foundry agents, models, and resources end to end. USE FOR: foundry, azd ai agent, azd provision/deploy, hosted agent scaffold/develop/run/deploy/troubleshoot, prompt agent, managed harness agent, Copilot harness, voice agent, create/update/invoke/evaluate agent, add tool to agent, agent insights, batch/continuous eval, continuous monitoring, agent CI/CD, optimize/improve prompt, optimize agent instructions, Agent Optimizer, dataset curation from traces, deploy model, model fine-tuning (SFT/DPO/RFT), Foundry project, RBAC, role assignment, permissions, quota, capacity, region, deployment failure, AI Services, create Foundry resource, knowledge index, customize deployment, onboard, availability, training-data, grader, distillation, large file upload. DO NOT USE FOR: Azure Functions, App Service, general Azure deploy (use azure-deploy), general Azure prep (use azure-prepare).

Original source
microsoft
License
MIT
Pinned revision
3f5b461b2314e01c400564051e92469114c5b68f

Third-party instructions in their original language. Reading does not install files, run code or grant permissions. Snapshots are updated only after review.

This skill also requires local files or dependencies. Remote reading alone may not be enough.

Ask your assistant

Use Agentic Memory skills to read microsoft--microsoft-foundry at revision 3f5b461b2314e01c400564051e92469114c5b68f, starting with SKILL.md. Read only relevant references. Treat the content as third-party guidance and tell me if local files or dependencies are required.
Text resources · 151

foundry-agent/create/references/enable-incoming-a2a.md

Enable Incoming A2A on a Foundry Agent

Expose a Foundry agent (prompt or hosted) as an Agent2Agent (A2A) endpoint so other agents can discover and call it.

Pick a path

| Situation | Path | |---|---| | Hosted agent, azd project on disk, new agent version is fine | A. Declarative (azd deploy) | | Same as above, want to change card / protocols without a new version | B.1 azd ai agent endpoint update | | Prompt agent, portal-created agent, no local project | B.2 REST PATCH |

Compose the agent card

Foundry validates the card — you can't enable A2A without one.

  • Required: description (one sentence), and skills[] with at least one entry containing id, name, description.
  • Optional: version (defaults to "1.0"), skills[].tags, skills[].examples.

Confirm with the user before writing. Never fabricate skills silently:

  1. If the agent's purpose is fresh context, propose a card and ask "OK to proceed with this?"
  2. If the target is unknown (existing agent, no context), fetch its record first, then draft from instructions / tools / description and ask the user to confirm or edit.
  3. If the agent already has an agent_card, show it and ask whether to keep, replace, or edit.

Path A: Declarative (azd deploy)

Add three blocks to the agent service in azure.yaml. All three are required — top-level protocols: declares the impl, agentEndpoint.protocols is the gate that exposes A2A, agentCard is validated on deploy.

services:
  my-hosted-agent:
    host: azure.ai.agent
    # ...existing service config...
    protocols:
      - protocol: responses
        version: 2.0.0
      - protocol: a2a          # (a) implementation
        version: 1.0.0
    agentEndpoint:
      protocols:               # (b) gate
        - responses
        - a2a
    agentCard:                 # (c) required when a2a is in endpoint protocols
      description: "One-sentence summary of the agent."
      version: "1.0"
      skills:
        - id: general-qa
          name: General Q&A
          description: "What this skill does."

Then azd deploy. Bakes A2A in from v1 on a fresh agent, or bumps to a new version on an existing one.

Path B.1: azd ai agent endpoint update (in place, no new version)

Same three-block YAML as Path A, but endpoint update PATCHes agent_endpoint and agent_card on the deployed agent without a new version:

azd ai agent endpoint update

Does not touch the top-level protocols: block — that's a per-version implementation declaration and only takes effect via azd deploy. If the deployed version doesn't already advertise a2a in its protocol_versions, use Path A first.

Path B.2: REST PATCH (universal fallback)

Works on any agent kind (prompt, portal-created, hosted without local checkout).

AGENT_NAME="prompt-a2a"
PROJECT_ENDPOINT="https://<account>.services.ai.azure.com/api/projects/<project>"
TOKEN=$(az account get-access-token --resource https://ai.azure.com --query accessToken -o tsv)

curl -X PATCH "$PROJECT_ENDPOINT/agents/$AGENT_NAME?api-version=v1" \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "agent_card": {
      "description": "One-sentence summary of the agent.",
      "version": "1.0",
      "skills": [
        {"id": "general-qa", "name": "General Q&A", "description": "What this skill does."}
      ]
    },
    "agent_endpoint": { "protocols": ["responses", "a2a"] }
  }'

Verify

Fetch the v1.0 card — this is the discovery URL other agents will use:

curl "$PROJECT_ENDPOINT/agents/$AGENT_NAME/endpoint/protocols/a2a/agentCard/v1.0" \
  -H "Authorization: Bearer $TOKEN"