fix(compression): skip auto-lower when context_length is user-configured

agent/conversation_compression.py

@@ -264,7 +264,14 @@ def check_compression_model_feasibility(agent: Any) -> None:
             )
 
         threshold = agent.context_compressor.threshold_tokens
-        if aux_context < threshold:
+        # Skip auto-lower when the aux context came from explicit user
+        # config (auxiliary.compression.context_length).  A user-configured
+        # context_length declares an *input budget* for the summarizer,
+        # not a capability limit of the compression model — auto-lowering
+        # the session threshold to match it causes near-useless compression
+        # loops on long sessions (issue #58407).
+        _from_config = getattr(agent, "_aux_compression_context_length_config", None) is not None
+        if aux_context < threshold and not _from_config:
             # Auto-correct: lower the live session threshold so
             # compression actually works this session.  The hard floor
             # above guarantees aux_context >= MINIMUM_CONTEXT_LENGTH,

tests/run_agent/test_compression_config_auto_lower.py

@@ -0,0 +1,116 @@
+"""Regression: auto-lower must skip when context_length is user-configured.
+
+Issue #58407: ``auxiliary.compression.context_length`` in config.yaml set an
+input budget for the summarizer, but the auto-lower logic treated it as a
+capability limit and lowered the session threshold — causing near-useless
+compression loops on long sessions.
+
+When the aux context value originates from the user's explicit config
+(``_aux_compression_context_length_config is not None``), the auto-lower
+should be skipped.
+"""
+
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+from run_agent import AIAgent
+from agent.context_compressor import ContextCompressor
+
+
+@pytest.fixture(autouse=True)
+def _stable_aux_provider_config():
+    with patch(
+        "agent.auxiliary_client._resolve_task_provider_model",
+        return_value=("auto", None, None, None, None),
+    ):
+        yield
+
+
+def _make_agent(
+    *,
+    compression_enabled: bool = True,
+    threshold_percent: float = 0.50,
+    main_context: int = 800_000,
+    aux_context_config: int | None = None,
+) -> AIAgent:
+    agent = AIAgent.__new__(AIAgent)
+    agent.model = "test-main-model"
+    agent.provider = "openrouter"
+    agent.base_url = "https://openrouter.ai/api/v1"
+    agent.api_key = "sk-test"
+    agent.api_mode = "chat_completions"
+    agent.quiet_mode = True
+    agent.log_prefix = ""
+    agent.compression_enabled = compression_enabled
+    agent._print_fn = None
+    agent.suppress_status_output = False
+    agent._stream_consumers = []
+    agent._executing_tools = False
+    agent._mute_post_response = False
+    agent.status_callback = None
+    agent.tool_progress_callback = None
+    agent._compression_warning = None
+    agent._aux_compression_context_length_config = aux_context_config
+    agent._custom_providers = []
+    agent.tools = []
+
+    compressor = MagicMock(spec=ContextCompressor)
+    compressor.context_length = main_context
+    compressor.threshold_tokens = int(main_context * threshold_percent)
+    agent.context_compressor = compressor
+
+    return agent
+
+
+@patch("agent.model_metadata.get_model_context_length", return_value=128_000)
+@patch("agent.auxiliary_client.get_text_auxiliary_client")
+def test_auto_lower_skipped_when_context_length_from_config(mock_get_client, mock_ctx_len):
+    """Issue #58407: user-configured context_length is an input budget,
+    not a capability limit.  Auto-lower must NOT engage."""
+    # Main model: 800K context, threshold at 50% = 400K
+    # Aux model: 128K context (returned by mock)
+    # Config: auxiliary.compression.context_length = 128000
+    agent = _make_agent(
+        main_context=800_000,
+        threshold_percent=0.50,
+        aux_context_config=128_000,
+    )
+    mock_client = MagicMock()
+    mock_client.base_url = "https://openrouter.ai/api/v1"
+    mock_client.api_key = "sk-aux"
+    mock_get_client.return_value = (mock_client, "deepseek/deepseek-v4-flash")
+
+    messages = []
+    agent._emit_status = lambda msg: messages.append(msg)
+
+    agent._check_compression_model_feasibility()
+
+    # Threshold must NOT be lowered — the 128K came from user config
+    assert agent.context_compressor.threshold_tokens == 400_000
+    # No "Auto-lowered" message
+    assert not any("Auto-lowered" in m for m in messages)
+
+
+@patch("agent.model_metadata.get_model_context_length", return_value=128_000)
+@patch("agent.auxiliary_client.get_text_auxiliary_client")
+def test_auto_lower_still_engages_when_no_config_override(mock_get_client, mock_ctx_len):
+    """Without explicit config, auto-lower should still work (existing behavior)."""
+    agent = _make_agent(
+        main_context=800_000,
+        threshold_percent=0.50,
+        aux_context_config=None,  # no user config
+    )
+    mock_client = MagicMock()
+    mock_client.base_url = "https://openrouter.ai/api/v1"
+    mock_client.api_key = "sk-aux"
+    mock_get_client.return_value = (mock_client, "deepseek/deepseek-v4-flash")
+
+    messages = []
+    agent._emit_status = lambda msg: messages.append(msg)
+
+    agent._check_compression_model_feasibility()
+
+    # Threshold IS lowered — aux model's detected limit is below threshold
+    assert agent.context_compressor.threshold_tokens == 128_000
+    assert any("Auto-lowered" in m for m in messages)