fix(data): strip only the key prefix when reading DynamoDB thread ids

backend/chainlit/data/dynamodb.py

@@ -177,8 +177,8 @@ async def delete_feedback(self, feedback_id: str) -> bool:
 
         # feedback id = THREAD#{thread_id}::STEP#{step_id}
         thread_id, step_id = feedback_id.split("::")
-        thread_id = thread_id.strip("THREAD#")
-        step_id = step_id.strip("STEP#")
+        thread_id = thread_id.removeprefix("THREAD#")
+        step_id = step_id.removeprefix("STEP#")
 
         self.client.update_item(
             TableName=self.table_name,
@@ -500,8 +500,8 @@ async def list_threads(
         for item in response["Items"]:
             deserialized_item: Dict[str, Any] = self._deserialize_item(item)
             thread = ThreadDict(  # type: ignore
-                id=deserialized_item["PK"].strip("THREAD#"),
-                createdAt=deserialized_item["UserThreadSK"].strip("TS#"),
+                id=deserialized_item["PK"].removeprefix("THREAD#"),
+                createdAt=deserialized_item["UserThreadSK"].removeprefix("TS#"),
                 name=deserialized_item["name"],
             )
             paginated_response.data.append(thread)

backend/tests/data/test_dynamodb.py

@@ -0,0 +1,57 @@
+"""Tests for the DynamoDB data layer's thread and step identifier handling.
+
+DynamoDB stores a thread as ``PK="THREAD#<thread_id>"`` and a step as
+``SK="STEP#<step_id>"``. Reading those identifiers back must strip the prefix
+only, never the characters the identifier happens to end with.
+"""
+
+from unittest.mock import Mock
+
+import pytest
+
+from chainlit.data.dynamodb import DynamoDBDataLayer
+from chainlit.types import Pagination, ThreadFilter
+
+
+def _make_layer(items=None):
+    client = Mock()
+    client.query.return_value = {"Items": items or []}
+    return DynamoDBDataLayer(table_name="test-table", client=client), client
+
+
+@pytest.mark.asyncio
+async def test_list_threads_keeps_thread_ids_ending_in_prefix_characters():
+    """A thread id ending in T/H/R/E/A/D keeps every character.
+
+    ``str.strip("THREAD#")`` treats its argument as a set of characters, so it
+    eats the id's own trailing letters; ``removeprefix`` removes the prefix only.
+    """
+    layer, _ = _make_layer(
+        [
+            {
+                "PK": {"S": "THREAD#DEMO"},
+                "UserThreadSK": {"S": "TS#2026-01-01T00:00:00"},
+                "name": {"S": "demo thread"},
+            }
+        ]
+    )
+
+    response = await layer.list_threads(
+        Pagination(first=10), ThreadFilter(userId="user-1")
+    )
+
+    assert [thread["id"] for thread in response.data] == ["DEMO"]
+    assert response.data[0]["createdAt"] == "2026-01-01T00:00:00"
+
+
+@pytest.mark.asyncio
+async def test_delete_feedback_builds_the_key_from_the_full_identifiers():
+    """The delete key must reuse the ids from the feedback id verbatim."""
+    layer, client = _make_layer()
+
+    await layer.delete_feedback("THREAD#CODE::STEP#SEED")
+
+    assert client.update_item.call_args.kwargs["Key"] == {
+        "PK": {"S": "THREAD#CODE"},
+        "SK": {"S": "STEP#SEED"},
+    }