feat: improve WebUI run recovery

tests/test_langgraph_dev_http.py

@@ -7,6 +7,7 @@ from __future__ import annotations
 
 from unittest.mock import patch
 
+from langchain_core.messages import AIMessage, HumanMessage
 from starlette.testclient import TestClient
 
 from EvoScientist.config import EvoScientistConfig
@@ -161,3 +162,143 @@ def test_ollama_discovery_skipped_when_base_url_absent():
         {"name": n, "model_id": m, "provider": p}
         for n, m, p in list_models_by_provider()
     ]
+
+
+def test_final_answer_extracts_latest_ai_text_blocks():
+    async def fake_metadata(_thread_id):
+        return {"updated_at": "2026-07-06T14:14:53+00:00"}
+
+    async def fake_messages(_thread_id):
+        return [
+            HumanMessage(content="question"),
+            AIMessage(content="old answer"),
+            AIMessage(
+                content=[
+                    {"type": "reasoning", "text": "internal"},
+                    {"type": "text", "text": "Part A"},
+                    {"type": "tool_use", "name": "search"},
+                    {"type": "output_text", "text": "Part B"},
+                ]
+            ),
+        ]
+
+    async def fake_runtime(_request, _thread_id):
+        return {
+            "found": True,
+            "complete": True,
+            "completed_at": "2026-07-06T14:15:00+00:00",
+        }
+
+    with (
+        patch(
+            "EvoScientist.langgraph_dev.http._get_thread_metadata_for_http",
+            new=fake_metadata,
+        ),
+        patch(
+            "EvoScientist.langgraph_dev.http._get_thread_messages_for_http",
+            new=fake_messages,
+        ),
+        patch(
+            "EvoScientist.langgraph_dev.http._read_thread_runtime_state",
+            new=fake_runtime,
+        ),
+    ):
+        resp = client.get("/api/threads/thread-1/final-answer")
+
+    assert resp.status_code == 200
+    assert resp.json() == {
+        "content": "Part A\n\nPart B",
+        "completed_at": "2026-07-06T14:15:00+00:00",
+        "complete": True,
+    }
+
+
+def test_final_answer_skips_tool_selection_json_text():
+    async def fake_metadata(_thread_id):
+        return {"updated_at": "2026-07-06T14:14:53+00:00"}
+
+    async def fake_messages(_thread_id):
+        return [
+            HumanMessage(content="question"),
+            AIMessage(content="stable answer"),
+            AIMessage(
+                content=(
+                    '{"tools":["search_papers","get_abstract"]}'
+                    '{"tools":["web_search_exa"]}'
+                )
+            ),
+        ]
+
+    async def fake_runtime(_request, _thread_id):
+        return {
+            "found": True,
+            "complete": True,
+            "completed_at": "2026-07-06T14:15:00+00:00",
+        }
+
+    with (
+        patch(
+            "EvoScientist.langgraph_dev.http._get_thread_metadata_for_http",
+            new=fake_metadata,
+        ),
+        patch(
+            "EvoScientist.langgraph_dev.http._get_thread_messages_for_http",
+            new=fake_messages,
+        ),
+        patch(
+            "EvoScientist.langgraph_dev.http._read_thread_runtime_state",
+            new=fake_runtime,
+        ),
+    ):
+        resp = client.get("/api/threads/thread-1/final-answer")
+
+    assert resp.status_code == 200
+    assert resp.json()["content"] == "stable answer"
+
+
+def test_final_answer_returns_404_for_unknown_thread():
+    async def fake_metadata(_thread_id):
+        return None
+
+    with patch(
+        "EvoScientist.langgraph_dev.http._get_thread_metadata_for_http",
+        new=fake_metadata,
+    ):
+        resp = client.get("/api/threads/missing/final-answer")
+
+    assert resp.status_code == 404
+    assert resp.json() == {"error": "thread not found"}
+
+
+def test_final_answer_does_not_mark_complete_when_runtime_state_fails():
+    async def fake_metadata(_thread_id):
+        return {"updated_at": "2026-07-06T14:14:53+00:00"}
+
+    async def fake_messages(_thread_id):
+        return [AIMessage(content="checkpoint answer")]
+
+    async def fake_runtime(_request, _thread_id):
+        raise RuntimeError("langgraph runtime unavailable")
+
+    with (
+        patch(
+            "EvoScientist.langgraph_dev.http._get_thread_metadata_for_http",
+            new=fake_metadata,
+        ),
+        patch(
+            "EvoScientist.langgraph_dev.http._get_thread_messages_for_http",
+            new=fake_messages,
+        ),
+        patch(
+            "EvoScientist.langgraph_dev.http._read_thread_runtime_state",
+            new=fake_runtime,
+        ),
+    ):
+        resp = client.get("/api/threads/thread-1/final-answer")
+
+    assert resp.status_code == 200
+    assert resp.json() == {
+        "content": "checkpoint answer",
+        "completed_at": None,
+        "complete": False,
+    }