Automated sync from private repo (2026-07-17)

migration/README.md

@@ -116,6 +116,9 @@ az role assignment create `
 
 # Cross-project: read from one project, write to another
 .\migrate-docker.ps1 --source-resource-id "<SOURCE_ID>" --resource-id "<TARGET_ID>"
+
+# Migrate agent definitions ONLY, skip files/vector stores (advanced — see below)
+.\migrate-docker.ps1 --resource-id "<ID>" --no-migrate-files
 ```
 
 ---
@@ -134,6 +137,62 @@ Real v2 agents are named (`agent.name = "MyAgent"`, `agent.version = "1"`) — `
 
 ---
 
+## Files & vector stores
+
+When an assistant uses **`file_search`** or **`code_interpreter`**, its uploaded files and vector stores live in the *source* project. Those IDs don't exist in the target project, so a naive copy would leave the migrated agent with broken references. The tool handles this for you.
+
+### It's automatic — no flag required
+
+File and vector-store migration is **on by default**. For every assistant it migrates, the tool:
+
+1. Reads the tool resources on the v1 assistant.
+2. **Downloads** each referenced file from the source project.
+3. **Re-uploads** it to the target project's **Foundry endpoint** (`{resource}.services.ai.azure.com`).
+4. **Recreates** each `file_search` vector store on the target and attaches the re-uploaded files.
+5. **Rewrites** the file IDs and vector-store IDs in the new v2 agent definition so everything resolves.
+
+You'll see it in the run output:
+
+```text
+📂 File migration: ENABLED (target Foundry endpoint: https://<resource>.services.ai.azure.com/api/projects/<project>)
+   📦 Migrating files for assistant asst_abc123
+   🔧 code_interpreter: 2 file(s) to migrate
+   📥 Downloaded sales.csv (10240B)
+   📤 Uploaded sales.csv -> assistant-file-...
+   🔍 file_search: 1 vector store(s) to migrate
+   🗄️  Created vector store vs_...
+   ✅ Vector store vs_... ready (3 file(s) ingested)
+```
+
+### What gets migrated (and what doesn't)
+
+| Tool on the v1 assistant | What's migrated | Read from |
+| --- | --- | --- |
+| `code_interpreter` | Each attached file | `tool_resources.code_interpreter.file_ids` |
+| `file_search` | Every file inside each vector store, plus a freshly created vector store | `tool_resources.file_search.vector_store_ids` |
+| Any other tool (bing, function, openapi, …) | Nothing — these carry no files | — |
+
+- **Only files referenced by a tool are copied.** Stray files in the project that no assistant points at are left alone.
+- **No file-type or size filtering by this tool.** Whatever the source returns is re-uploaded as-is; the target Files / `file_search` API enforces the actual limits — currently a 512 MB max file size, 5,000,000 max tokens per file, and [these supported file types](https://learn.microsoft.com/azure/foundry/agents/how-to/tools/file-search#supported-file-types) (`.c`, `.cpp`, `.cs`, `.doc`/`.docx`, `.html`, `.java`, `.json`, `.md`, `.pdf`, `.php`, `.pptx`, `.py`, `.rb`, `.tex`, `.txt`, `.css`, `.js`, `.sh`, `.ts` — text files must be UTF-8/UTF-16/ASCII). A vector store can also hold at most 10,000 files. If a file is too large or an unsupported type, the API rejects it (HTTP 400/415, or the vector-store file status ends up `failed`) — the tool does **not** retry these, and reports the file id + API-provided reason (e.g. `[invalid_file] ...`) in the terminal, the assistant's `failed_files` list, and `migration_failed_files.json`.
+- **Files shared** between `code_interpreter` and a vector store are uploaded once and reused (deduplicated).
+
+### Turning it OFF — `--no-migrate-files`
+
+Pass `--no-migrate-files` to copy **agent definitions only**. The v2 agent keeps the *source* file / vector-store IDs, which do **not** exist on the target — so `file_search` and `code_interpreter` stay broken until you re-upload manually. Use this only when the target already has the files, or when you plan to fix references yourself.
+
+```powershell
+.\migrate-docker.ps1 --resource-id "<ID>" --no-migrate-files
+```
+
+### Requirements & caveats
+
+- The tool needs a reachable **source endpoint** and a derivable **target Foundry endpoint** (derived automatically from `--resource-id`). If either is missing it prints `⚠️  Cannot migrate files: …` and continues with definitions only.
+- **A wrong `--production-endpoint` is accepted, not rejected.** If you override the target with a raw `--production-endpoint` that points at a legacy OpenAI-compatible host (`*.openai.azure.com`) or a raw Cognitive Services host (`*.cognitiveservices.azure.com`) instead of a Foundry host (`*.services.ai.azure.com`), the migration **will still run and the file/vector-store upload calls will still succeed (HTTP 200)** — but the uploaded files land in a namespace the Foundry agent runtime can't see, so `file_search`/`code_interpreter` stay broken on the migrated agent. The tool cannot tell this happened from the API responses alone, so it proactively checks the endpoint's hostname and prints a `⚠️` warning up front and again in the final summary — it does **not** fail the migration, since you may not care about files at all (see `--no-migrate-files`). Re-run with a Foundry-format endpoint to actually get working files.
+- **Vector-store ingestion is polled per file, not per store.** Each file is attached to the vector store and polled individually (default: 30 attempts × 2s = up to 60s per file; configurable via `MIGRATION_VS_FILE_POLL_ATTEMPTS` / `MIGRATION_VS_FILE_POLL_INTERVAL`) until it reaches `completed` or `failed`. This isolates one bad file from the rest — the vector store is still created and attached with whatever files succeeded, even if the poll window elapses on a slow file (it keeps indexing server-side) or a file permanently fails.
+- You need write access (**Foundry User**) on the *target* resource — the same role required for the rest of the migration.
+
+---
+
 ## Don't have Docker? Use migrate.ps1 instead
 
 If you already have **Python 3.10+** and **Azure CLI** installed locally, you can skip Docker:

migration/v1_to_v2_migration.py

@@ -146,6 +146,19 @@ def get_target_openai_endpoint(production_resource: Optional[str] = None, produc
     return None
 
 
+def _is_foundry_host(url: Optional[str]) -> bool:
+    """
+    Return True if *url*'s hostname is a Foundry project endpoint
+    (``*.services.ai.azure.com``). Files/vector stores uploaded anywhere
+    else (legacy ``*.openai.azure.com`` / ``*.cognitiveservices.azure.com``)
+    are invisible to the Foundry agent runtime.
+    """
+    if not url:
+        return False
+    host = (urlparse(url).hostname or "").lower()
+    return host == "services.ai.azure.com" or host.endswith(".services.ai.azure.com")
+
+
 def get_target_foundry_endpoint(production_resource: Optional[str] = None, production_endpoint: Optional[str] = None) -> Optional[str]:
     """
     Return the Foundry endpoint for file uploads and vector-store creation.
@@ -3689,6 +3702,30 @@ def _has_tools(assistant: Dict[str, Any]) -> bool:
         production_endpoint=production_endpoint,
     )
 
+    # Although a target endpoint using the legacy OpenAI-compatible host
+    # (*.openai.azure.com) or the raw Cognitive Services host
+    # (*.cognitiveservices.azure.com) WILL BE ACCEPTED for migration — those
+    # hosts return HTTP 200 for the file/vector-store upload calls, so nothing
+    # in the migration "fails" — it will NOT actually migrate files/vector
+    # stores: the uploads land in a namespace the Foundry agent runtime can't
+    # see, so file_search/code_interpreter stay broken on the migrated agent.
+    # This wrong-but-accepted endpoint case is called out as a warning during
+    # migration (printed up front and repeated in the final summary) rather
+    # than treated as fatal, since callers who passed --no-migrate-files don't
+    # care about this at all.
+    file_migration_endpoint_warning: Optional[str] = None
+    if migrate_files and target_v2_endpoint and not _is_foundry_host(target_v2_endpoint):
+        file_migration_endpoint_warning = (
+            f"--production-endpoint ({target_v2_endpoint}) is a legacy OpenAI-compatible "
+            "(*.openai.azure.com) or Cognitive Services (*.cognitiveservices.azure.com) host, "
+            "not a Foundry (*.services.ai.azure.com) host. This endpoint WILL BE ACCEPTED for "
+            "migration — the upload calls succeed — but it will NOT migrate files/vector stores; "
+            "they end up invisible to the Foundry agent runtime. Re-run with a Foundry-format "
+            "--production-endpoint (or --production-resource) to migrate files successfully, "
+            "or pass --no-migrate-files if you don't need files migrated."
+        )
+        print(f"\n⚠️  {file_migration_endpoint_warning}\n")
+
     # In-place upgrade (source == target): when no explicit target was supplied
     # but the source project is itself a Foundry endpoint, reuse it as the target
     # so the duplicate-migration guard still runs. Without this, re-running the
@@ -4003,6 +4040,9 @@ def ensure_tools_array():
     # Always using v2 API
     print(f"   Target: v2 API ({BASE_V2})")
 
+    if file_migration_endpoint_warning:
+        print(f"\n⚠️  File/vector-store migration warning: {file_migration_endpoint_warning}")
+
 
 def _classify_v1_item(item: Dict[str, Any]) -> str:
     """

samples/csharp/hosted-agents/agent-framework/browser-automation/azure.yaml

@@ -3,7 +3,7 @@
 requiredVersions:
   extensions:
     azure.ai.agents: '>=1.0.0-beta.4'
-name: browser-automation-csharp-maf-sample-foundry
+name: bat-csharp-maf
 services:
   ai-project:
     host: azure.ai.project
@@ -35,7 +35,7 @@ services:
         browser_automation_preview:
           connection:
             project_connection_id: '${AZURE_AI_PROJECT_ID}/connections/browserautomation'
-  browser-automation-csharp-maf-sample-foundry:
+  bat-csharp-maf:
     host: azure.ai.agent
     metadata:
       tags:
@@ -44,7 +44,7 @@ services:
         - Azure AI AgentServer
         - Responses Protocol
         - Browser Automation
-    project: src/browser-automation-csharp-maf-sample-foundry
+    project: src/bat-csharp-maf
     language: csharp
     docker:
       path: Dockerfile
@@ -53,7 +53,7 @@ services:
       - browserautomation
       - browser-automation-tools
     kind: hosted
-    name: browser-automation-csharp-maf-sample-foundry
+    name: bat-csharp-maf
     displayName: Browser Automation Agent (C#, MAF)
     description: A Foundry-hosted browser automation agent (C#, MAF) for general browsing, web scraping, and form filling.
     protocols:

samples/csharp/hosted-agents/bring-your-own/responses/browser-automation/azure.yaml

@@ -3,7 +3,7 @@
 requiredVersions:
   extensions:
     azure.ai.agents: '>=1.0.0-beta.4'
-name: browser-automation-csharp-byo-sample-foundry
+name: bat-csharp-byo
 services:
   ai-project:
     host: azure.ai.project
@@ -35,7 +35,7 @@ services:
         browser_automation_preview:
           connection:
             project_connection_id: '${AZURE_AI_PROJECT_ID}/connections/browserautomation'
-  browser-automation-csharp-byo-sample-foundry:
+  bat-csharp-byo:
     host: azure.ai.agent
     metadata:
       tags:
@@ -44,7 +44,7 @@ services:
         - Bring Your Own
         - Browser Automation
         - .NET
-    project: src/browser-automation-csharp-byo-sample-foundry
+    project: src/bat-csharp-byo
     language: csharp
     docker:
       path: Dockerfile
@@ -53,7 +53,7 @@ services:
       - browserautomation
       - browser-automation-tools
     kind: hosted
-    name: browser-automation-csharp-byo-sample-foundry
+    name: bat-csharp-byo
     displayName: Browser Automation Agent (C#, BYO Responses)
     description: A Foundry-hosted browser automation agent (C#, BYO Responses) for general browsing, web scraping, and form filling.
     protocols:

samples/python/hosted-agents/agent-framework/responses/14-browser-automation-agent/azure.yaml

@@ -3,7 +3,7 @@
 requiredVersions:
   extensions:
     azure.ai.agents: '>=1.0.0-beta.4'
-name: browser-automation-python-maf-sample-foundry
+name: bat-python-maf
 services:
   ai-project:
     host: azure.ai.project
@@ -35,7 +35,7 @@ services:
         browser_automation_preview:
           connection:
             project_connection_id: '${AZURE_AI_PROJECT_ID}/connections/browserautomation'
-  browser-automation-python-maf-sample-foundry:
+  bat-python-maf:
     host: azure.ai.agent
     metadata:
       tags:
@@ -44,7 +44,7 @@ services:
         - Azure AI AgentServer
         - Responses Protocol
         - Browser Automation
-    project: src/browser-automation-python-maf-sample-foundry
+    project: src/bat-python-maf
     language: python
     docker:
       path: Dockerfile
@@ -53,7 +53,7 @@ services:
       - browserautomation
       - browser-automation-tools
     kind: hosted
-    name: browser-automation-python-maf-sample-foundry
+    name: bat-python-maf
     description: A Foundry-hosted browser automation agent sample for general browsing, web scraping, and form filling.
     protocols:
       - protocol: responses

samples/python/hosted-agents/bring-your-own/responses/browser-automation/azure.yaml

@@ -3,7 +3,7 @@
 requiredVersions:
   extensions:
     azure.ai.agents: '>=1.0.0-beta.4'
-name: browser-automation-python-byo-sample-foundry
+name: bat-python-byo
 services:
   ai-project:
     host: azure.ai.project
@@ -35,7 +35,7 @@ services:
         browser_automation_preview:
           connection:
             project_connection_id: '${AZURE_AI_PROJECT_ID}/connections/browserautomation'
-  browser-automation-python-byo-sample-foundry:
+  bat-python-byo:
     host: azure.ai.agent
     metadata:
       tags:
@@ -44,7 +44,7 @@ services:
         - Azure AI AgentServer
         - Responses Protocol
         - Browser Automation
-    project: src/browser-automation-python-byo-sample-foundry
+    project: src/bat-python-byo
     language: python
     docker:
       path: Dockerfile
@@ -53,7 +53,7 @@ services:
       - browserautomation
       - browser-automation-tools
     kind: hosted
-    name: browser-automation-python-byo-sample-foundry
+    name: bat-python-byo
     description: A Foundry-hosted browser automation agent sample for general browsing, web scraping, and form filling.
     protocols:
       - protocol: responses