fix: linting and typechecks

Author: GangGreenTemperTatum

dreadnode/agent/hooks/summarize.py

@@ -76,10 +76,12 @@ def _find_tool_aware_boundary(
         has_orphan = False
         for msg in messages[boundary:]:
             if msg.role == "tool" and hasattr(msg, "tool_call_id"):
-                call_idx = tool_call_map.get(msg.tool_call_id)
-                if call_idx is not None and call_idx < boundary:
-                    has_orphan = True
-                    break
+                tool_call_id = getattr(msg, "tool_call_id", None)
+                if tool_call_id is not None:
+                    call_idx = tool_call_map.get(tool_call_id)
+                    if call_idx is not None and call_idx < boundary:
+                        has_orphan = True
+                        break
 
         if not has_orphan:
             return boundary
@@ -93,6 +95,7 @@ def summarize_when_long(
     max_tokens: int = 100_000,
     min_messages_to_keep: int = 5,
     guidance: str = "",
+    *,
     preserve_tool_pairs: bool = True,
 ) -> "Hook":
     """

tests/test_preserve_tool_pairs.py

@@ -1,27 +1,41 @@
 """Tests for preserve_tool_pairs functionality in summarize_when_long hook."""
 
 import rigging as rg
+
 from dreadnode.agent.hooks.summarize import _find_tool_aware_boundary
 
 
 def test_preserves_tool_pairs():
-    """Tool call and response stay together when split."""
+    """Tool call and response stay together when boundary would split them."""
     messages = [
         rg.Message("user", "Hello"),
         rg.Message(
             "assistant",
             "Let me check",
-            tool_calls=[{"id": "call_1", "type": "function", "function": {"name": "check", "arguments": "{}"}}],
+            tool_calls=[
+                {
+                    "id": "call_1",
+                    "type": "function",
+                    "function": {"name": "check", "arguments": "{}"},
+                }
+            ],
         ),
         rg.Message("tool", "Result", tool_call_id="call_1"),
         rg.Message("assistant", "Done"),
         rg.Message("user", "Thanks"),
     ]
 
-    boundary = _find_tool_aware_boundary(messages, min_messages_to_keep=2)
+    # With min=3, naive boundary would be at index 2, keeping [2,3,4]
+    # But that would orphan the tool response at index 2 (call at index 1)
+    # So boundary should move back to index 1 to keep the pair together
+    boundary = _find_tool_aware_boundary(messages, min_messages_to_keep=3)
+    assert boundary == 1, f"Expected boundary 1 to keep tool pair together, got {boundary}"
 
-    # Should keep tool pair together by moving boundary earlier
-    assert boundary <= 1, "Boundary should preserve tool call/response pair"
+    # Verify the kept messages include the complete tool pair
+    kept = messages[boundary:]
+    assert len(kept) == 4
+    assert kept[0].role == "assistant"
+    assert kept[1].role == "tool"
 
 
 def test_no_tools():
@@ -44,49 +58,59 @@ def test_multiple_tool_pairs():
         rg.Message(
             "assistant",
             "Running A",
-            tool_calls=[{"id": "a", "type": "function", "function": {"name": "run_a", "arguments": "{}"}}],
+            tool_calls=[
+                {"id": "a", "type": "function", "function": {"name": "run_a", "arguments": "{}"}}
+            ],
         ),
         rg.Message("tool", "A done", tool_call_id="a"),
         rg.Message(
             "assistant",
             "Running B",
-            tool_calls=[{"id": "b", "type": "function", "function": {"name": "run_b", "arguments": "{}"}}],
+            tool_calls=[
+                {"id": "b", "type": "function", "function": {"name": "run_b", "arguments": "{}"}}
+            ],
         ),
         rg.Message("tool", "B done", tool_call_id="b"),
         rg.Message("user", "Thanks"),
     ]
 
-    boundary = _find_tool_aware_boundary(messages, min_messages_to_keep=2)
-
-    # Should not split between any tool pairs
-    kept = messages[boundary:]
-    assert len(kept) >= 2, "Should keep minimum messages"
+    # With min=3, naive boundary at index 3 would keep [3,4,5]
+    # But index 4 (tool response "b") references call at index 3
+    # So boundary should move back to index 3 to keep the second pair
+    boundary = _find_tool_aware_boundary(messages, min_messages_to_keep=3)
+    assert boundary == 3, f"Expected boundary 3 to preserve second tool pair, got {boundary}"
 
 
 def test_no_valid_boundary():
-    """Returns 0 when entire conversation is tool chain."""
+    """Returns 0 when min_messages would force splitting all tool pairs."""
     messages = [
         rg.Message(
             "assistant",
             "Start",
-            tool_calls=[{"id": "1", "type": "function", "function": {"name": "start", "arguments": "{}"}}],
+            tool_calls=[
+                {"id": "1", "type": "function", "function": {"name": "start", "arguments": "{}"}}
+            ],
         ),
         rg.Message("tool", "Result 1", tool_call_id="1"),
         rg.Message(
             "assistant",
             "Continue",
-            tool_calls=[{"id": "2", "type": "function", "function": {"name": "continue", "arguments": "{}"}}],
+            tool_calls=[
+                {"id": "2", "type": "function", "function": {"name": "continue", "arguments": "{}"}}
+            ],
         ),
         rg.Message("tool", "Result 2", tool_call_id="2"),
     ]
 
-    boundary = _find_tool_aware_boundary(messages, min_messages_to_keep=2)
-    assert boundary == 0, "Should keep everything when no valid split exists"
+    # With min=3, we'd need to keep last 3 messages
+    # Any boundary would orphan at least one tool response
+    # So should return 0 to keep everything
+    boundary = _find_tool_aware_boundary(messages, min_messages_to_keep=3)
+    assert boundary == 0, f"Should keep everything when no valid split exists, got {boundary}"
 
 
 if __name__ == "__main__":
     test_preserves_tool_pairs()
     test_no_tools()
     test_multiple_tool_pairs()
     test_no_valid_boundary()
-    print("All tests passed")