Foundry Agent Sync

Overview

Create and synchronize prompt-based AI agents directly within Azure AI Foundry via the Agent Service REST API. This skill registers agents in the Foundry service itself — making them immediately available for invocation, evaluation, and management through the Foundry portal or API. Each agent is created or updated idempotently via a named POST call, using definitions from a local JSON manifest file.

Key distinction: This skill creates agents inside AI Foundry (server-side). It does not scaffold local agent code or container images — for that, use the microsoft-foundry skill's create sub-skill.

Prerequisites

The user must have:

1. An Azure AI Foundry project with a deployed model (e.g. gpt-5-4)
2. Azure CLI (az) authenticated with access to the Foundry project
3. The Azure AI User role (or higher) on the Foundry project resource

Collect these values before proceeding:

Manifest Format

The manifest is a JSON array where each entry defines one agent. Look for it at common paths: infra/foundry-agents.json, foundry-agents.json, or .foundry/agents.json. If none exists, scaffold one.

[
  {
    "useCaseId": "alert-triage",
    "description": "Short description of what this agent does.",
    "baseInstruction": "You are an assistant that... <system prompt for the agent>"
  }
]

Field Reference

Sync Script

PowerShell (interactive / CI)

Create or locate the sync script. The canonical path is infra/scripts/sync-foundry-agents.ps1 but adapt to the repo layout.

param(
  [Parameter(Mandatory)]
  [string]$SubscriptionId,

  [Parameter(Mandatory)]
  [string]$ProjectEndpoint,

  [string]$ManifestPath = (Join-Path $PSScriptRoot '..\foundry-agents.json'),
  [string]$ModelName = 'gpt-5-4',
  [string]$AgentNamePrefix = 'myproject',
  [string]$ApiVersion = '2025-11-15-preview'
)

$ErrorActionPreference = 'Stop'

# Optional: append a common instruction suffix to every agent
$commonSuffix = ''

az account set --subscription $SubscriptionId | Out-Null
$accessToken = az account get-access-token --resource https://ai.azure.com/ --query accessToken -o tsv
if (-not $accessToken) { throw 'Failed to acquire Foundry access token.' }

$definitions = Get-Content -Raw -Path $ManifestPath | ConvertFrom-Json
$headers = @{ Authorization = "Bearer $accessToken" }
$results = @()

foreach ($def in $definitions) {
  $agentName = "$AgentNamePrefix-$($def.useCaseId)"
  $instructions = if ($commonSuffix) { "$($def.baseInstruction)`n`n$commonSuffix" } else { $def.baseInstruction }
  $body = @{
    definition  = @{ kind = 'prompt'; model = $ModelName; instructions = $instructions }
    description = $def.description
    metadata    = @{ useCaseId = $def.useCaseId; managedBy = 'foundry-agent-sync' }
  } | ConvertTo-Json -Depth 8

  $uri = "$($ProjectEndpoint.TrimEnd('/'))/agents/$agentName`?api-version=$ApiVersion"
  $resp = Invoke-RestMethod -Method Post -Uri $uri -Headers $headers -ContentType 'application/json' -Body $body
  $version = $resp.version ?? $resp.latest_version ?? $resp.id ?? 'unknown'
  Write-Host "Synced $agentName ($version)"
  $results += [pscustomobject]@{ name = $agentName; version = $version }
}

$results | Format-Table -AutoSize

Bash (Bicep deployment script / CI)

For automated deployment via Microsoft.Resources/deploymentScripts, use a bash script that:

1. Authenticates with a managed identity: az login --identity --username "$CLIENT_ID"
2. Acquires a Foundry token: az account get-access-token --resource https://ai.azure.com/
3. Iterates definitions from the FOUNDRY_AGENT_DEFINITIONS environment variable (JSON string)
4. POSTs each agent to {endpoint}/agents/{name}?api-version=2025-11-15-preview

Bicep Integration (optional)

To run the sync automatically during infrastructure deployment:

…