[loop-cycle-61] fix: strip think tags and harden fact parsing (#237) (#254)

tests/timmy/test_thinking.py

@@ -885,3 +885,98 @@ async def test_call_agent_does_not_use_session_chat(tmp_path):
         await engine._call_agent("prompt")
 
     mock_session_chat.assert_not_awaited()
+
+
+@pytest.mark.asyncio
+async def test_call_agent_strips_think_tags(tmp_path):
+    """_call_agent must strip <think> tags from reasoning models like qwen3."""
+    engine = _make_engine(tmp_path)
+
+    mock_agent = AsyncMock()
+    mock_run = AsyncMock()
+    mock_run.content = (
+        "<think>Let me reason about this carefully...</think>"
+        "The actual thought content."
+    )
+    mock_agent.arun.return_value = mock_run
+
+    with patch("timmy.agent.create_timmy", return_value=mock_agent):
+        result = await engine._call_agent("test prompt")
+
+    assert "<think>" not in result
+    assert result == "The actual thought content."
+
+
+@pytest.mark.asyncio
+async def test_call_agent_strips_multiline_think_tags(tmp_path):
+    """_call_agent handles multi-line <think> blocks."""
+    engine = _make_engine(tmp_path)
+
+    mock_agent = AsyncMock()
+    mock_run = AsyncMock()
+    mock_run.content = (
+        "<think>\nStep 1: analyze\nStep 2: synthesize\n</think>\n"
+        "Clean output here."
+    )
+    mock_agent.arun.return_value = mock_run
+
+    with patch("timmy.agent.create_timmy", return_value=mock_agent):
+        result = await engine._call_agent("test prompt")
+
+    assert "<think>" not in result
+    assert result == "Clean output here."
+
+
+# ---------------------------------------------------------------------------
+# _parse_facts_response resilience (#237)
+# ---------------------------------------------------------------------------
+
+
+def test_parse_facts_clean_json(tmp_path):
+    """Direct JSON array should parse normally."""
+    engine = _make_engine(tmp_path)
+    result = engine._parse_facts_response('["fact one", "fact two"]')
+    assert result == ["fact one", "fact two"]
+
+
+def test_parse_facts_empty_array(tmp_path):
+    """Empty JSON array should return empty list."""
+    engine = _make_engine(tmp_path)
+    assert engine._parse_facts_response("[]") == []
+
+
+def test_parse_facts_with_prose_prefix(tmp_path):
+    """JSON array preceded by prose should still parse (#237)."""
+    engine = _make_engine(tmp_path)
+    raw = 'Here are the facts:\n["Alexander prefers YAML", "Timmy runs locally"]'
+    result = engine._parse_facts_response(raw)
+    assert result == ["Alexander prefers YAML", "Timmy runs locally"]
+
+
+def test_parse_facts_with_markdown_fences(tmp_path):
+    """JSON wrapped in markdown code fences should parse."""
+    engine = _make_engine(tmp_path)
+    raw = '```json\n["fact in fences"]\n```'
+    result = engine._parse_facts_response(raw)
+    assert result == ["fact in fences"]
+
+
+def test_parse_facts_filters_non_strings(tmp_path):
+    """Non-string entries in the array should be filtered out."""
+    engine = _make_engine(tmp_path)
+    result = engine._parse_facts_response('[42, "valid fact", null, true]')
+    assert result == ["valid fact"]
+
+
+def test_parse_facts_none_and_empty(tmp_path):
+    """None and empty input should return empty list."""
+    engine = _make_engine(tmp_path)
+    assert engine._parse_facts_response(None) == []
+    assert engine._parse_facts_response("") == []
+    assert engine._parse_facts_response("   ") == []
+
+
+def test_parse_facts_invalid_json(tmp_path):
+    """Totally invalid text with no JSON array should return empty list."""
+    engine = _make_engine(tmp_path)
+    assert engine._parse_facts_response("no json here at all") == []