Make asset mime_type immutable after first ingest

Prevents cross-tenant metadata mutation when multiple references share
the same content-addressed Asset row. mime_type can now only be set when
NULL (first ingest); subsequent attempts to change it are silently ignored.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

app/assets/database/queries/asset.py

@@ -69,7 +69,7 @@ def upsert_asset(
         if asset.size_bytes != int(size_bytes) and int(size_bytes) > 0:
             asset.size_bytes = int(size_bytes)
             changed = True
-        if mime_type and asset.mime_type != mime_type:
+        if mime_type and not asset.mime_type:
             asset.mime_type = mime_type
             changed = True
         if changed:
@@ -118,7 +118,7 @@ def update_asset_hash_and_mime(
         return False
     if asset_hash is not None:
         asset.hash = asset_hash
-    if mime_type is not None:
+    if mime_type is not None and not asset.mime_type:
         asset.mime_type = mime_type
     return True
 

app/assets/services/asset_management.py

@@ -108,10 +108,11 @@ def update_asset_metadata(
             touched = True
 
         if mime_type is not None:
-            update_asset_hash_and_mime(
+            updated = update_asset_hash_and_mime(
                 session, asset_id=ref.asset_id, mime_type=mime_type
             )
-            touched = True
+            if updated:
+                touched = True
 
         if preview_id is not None:
             set_reference_preview(

app/assets/services/ingest.py

@@ -146,7 +146,7 @@ def _register_existing_asset(
         if not asset:
             raise ValueError(f"No asset with hash {asset_hash}")
 
-        if mime_type and asset.mime_type != mime_type:
+        if mime_type and not asset.mime_type:
             update_asset_hash_and_mime(session, asset_id=asset.id, mime_type=mime_type)
 
         if preview_id:

tests-unit/assets_test/queries/test_asset.py

@@ -10,6 +10,7 @@ from app.assets.database.queries import (
     get_asset_by_hash,
     upsert_asset,
     bulk_insert_assets,
+    update_asset_hash_and_mime,
 )
 
 
@@ -142,3 +143,45 @@ class TestBulkInsertAssets:
         session.commit()
 
         assert session.query(Asset).count() == 200
+
+
+class TestMimeTypeImmutability:
+    """mime_type on Asset is write-once: set on first ingest, never overwritten."""
+
+    @pytest.mark.parametrize(
+        "initial_mime,second_mime,expected_mime",
+        [
+            ("image/png", "image/jpeg", "image/png"),
+            (None, "image/png", "image/png"),
+        ],
+        ids=["preserves_existing", "fills_null"],
+    )
+    def test_upsert_mime_immutability(self, session: Session, initial_mime, second_mime, expected_mime):
+        h = f"blake3:upsert_{initial_mime}_{second_mime}"
+        upsert_asset(session, asset_hash=h, size_bytes=100, mime_type=initial_mime)
+        session.commit()
+
+        asset, created, _ = upsert_asset(session, asset_hash=h, size_bytes=100, mime_type=second_mime)
+        assert created is False
+        assert asset.mime_type == expected_mime
+
+    @pytest.mark.parametrize(
+        "initial_mime,update_mime,update_hash,expected_mime,expected_hash",
+        [
+            (None, "image/png", None, "image/png", "blake3:upd0"),
+            ("image/png", "image/jpeg", None, "image/png", "blake3:upd1"),
+            ("image/png", "image/jpeg", "blake3:upd2_new", "image/png", "blake3:upd2_new"),
+        ],
+        ids=["fills_null", "preserves_existing", "hash_updates_mime_locked"],
+    )
+    def test_update_asset_hash_and_mime_immutability(
+        self, session: Session, initial_mime, update_mime, update_hash, expected_mime, expected_hash,
+    ):
+        h = expected_hash.removesuffix("_new")
+        asset = Asset(hash=h, size_bytes=100, mime_type=initial_mime)
+        session.add(asset)
+        session.flush()
+
+        update_asset_hash_and_mime(session, asset_id=asset.id, mime_type=update_mime, asset_hash=update_hash)
+        assert asset.mime_type == expected_mime
+        assert asset.hash == expected_hash