feat(conductor): Expose history bleed flags

This change introduces a new function, get_history_bleed_stats, to calculate and expose how close the current conversation history is to the provider's token limit. The initial implementation supports Anthropic, with a placeholder for Gemini.

ai_client.py

@@ -1146,3 +1146,37 @@ def send(
     elif _provider == "anthropic":
         return _send_anthropic(md_content, user_message, base_dir, file_items)
     raise ValueError(f"unknown provider: {_provider}")
+
+def get_history_bleed_stats() -> dict:
+    """
+    Calculates how close the current conversation history is to the token limit.
+    """
+    if _provider == "anthropic":
+        # For Anthropic, we have a robust estimator
+        current_tokens = _estimate_prompt_tokens([], _anthropic_history)
+        limit_tokens = _ANTHROPIC_MAX_PROMPT_TOKENS
+        percentage = (current_tokens / limit_tokens) * 100 if limit_tokens > 0 else 0
+        return {
+            "provider": "anthropic",
+            "limit": limit_tokens,
+            "current": current_tokens,
+            "percentage": percentage,
+        }
+    elif _provider == "gemini":
+        # For Gemini, token estimation is complex and handled by the server.
+        # We don't have a reliable client-side estimate, so we return a
+        # "not implemented" state for now.
+        return {
+            "provider": "gemini",
+            "limit": _GEMINI_MAX_INPUT_TOKENS,
+            "current": 0,
+            "percentage": 0,
+        }
+    
+    # Default empty state
+    return {
+        "provider": _provider,
+        "limit": 0,
+        "current": 0,
+        "percentage": 0,
+    }

tests/test_history_bleed.py

@@ -0,0 +1,56 @@
+import pytest
+from unittest.mock import patch, MagicMock
+
+# Import the module to be tested
+import ai_client
+
+@pytest.fixture(autouse=True)
+def reset_ai_client_session():
+    """Fixture to automatically reset the ai_client session before each test."""
+    ai_client.reset_session()
+
+def test_anthropic_history_bleed_calculation():
+    """
+    Tests that get_history_bleed_stats calculates the token usage
+    percentage correctly for the Anthropic provider.
+    """
+    # 1. Set up the test environment
+    ai_client.set_provider("anthropic", "claude-3-opus-20240229")
+
+    # Define the mock return value for the token estimator
+    mock_token_count = 150_000
+    # The hardcoded limit in the module is 180_000
+    expected_percentage = (mock_token_count / 180_000) * 100
+
+    # 2. Mock the internal dependencies
+    # We patch _estimate_prompt_tokens as it's the core of the calculation for anthropic
+    with patch('ai_client._estimate_prompt_tokens', return_value=mock_token_count) as mock_estimator:
+        
+        # 3. Call the function under test (which doesn't exist yet)
+        stats = ai_client.get_history_bleed_stats()
+
+        # 4. Assert the results
+        assert stats["provider"] == "anthropic"
+        assert stats["limit"] == 180_000
+        assert stats["current"] == mock_token_count
+        assert stats["percentage"] == pytest.approx(expected_percentage)
+        
+        # Ensure the mock was called
+        mock_estimator.assert_called_once()
+
+def test_gemini_history_bleed_not_implemented():
+    """
+    Tests that get_history_bleed_stats returns a 'not implemented' state
+    for Gemini, as its token calculation is different.
+    """
+    # 1. Set up the test environment
+    ai_client.set_provider("gemini", "gemini-1.5-pro-latest")
+
+    # 2. Call the function
+    stats = ai_client.get_history_bleed_stats()
+
+    # 3. Assert the 'not implemented' state
+    assert stats["provider"] == "gemini"
+    assert stats["limit"] == 900_000 # The constant _GEMINI_MAX_INPUT_TOKENS
+    assert stats["current"] == 0
+    assert stats["percentage"] == 0