fix(compressor): pass threshold_percent through update_model on model switch

agent/context_compressor.py

@@ -353,6 +353,7 @@ def update_model(
         api_key: str = "",
         provider: str = "",
         api_mode: str = "",
+        threshold_percent: float | None = None,
     ) -> None:
         """Update model info after a model switch or fallback activation."""
         self.model = model
@@ -361,6 +362,8 @@ def update_model(
         self.provider = provider
         self.api_mode = api_mode
         self.context_length = context_length
+        if threshold_percent is not None:
+            self.threshold_percent = threshold_percent
         self.threshold_tokens = max(
             int(context_length * self.threshold_percent),
             MINIMUM_CONTEXT_LENGTH,

run_agent.py

@@ -2372,6 +2372,7 @@ def switch_model(self, new_model, new_provider, api_key='', base_url='', api_mod
                 api_key=getattr(self, "api_key", ""),
                 provider=self.provider,
                 api_mode=self.api_mode,
+                threshold_percent=self.context_compressor.threshold_percent,
             )
 
         # ── Invalidate cached system prompt so it rebuilds next turn ──
@@ -2394,6 +2395,7 @@ def switch_model(self, new_model, new_provider, api_key='', base_url='', api_mod
             "compressor_provider": getattr(_cc, "provider", self.provider) if _cc else self.provider,
             "compressor_context_length": _cc.context_length if _cc else 0,
             "compressor_threshold_tokens": _cc.threshold_tokens if _cc else 0,
+            "compressor_threshold_percent": _cc.threshold_percent if _cc else 0.50,
         }
         if api_mode == "anthropic_messages":
             self._primary_runtime.update({
@@ -7611,6 +7613,7 @@ def _try_activate_fallback(self, reason: "FailoverReason | None" = None) -> bool
                     base_url=self.base_url,
                     api_key=getattr(self, "api_key", ""),
                     provider=self.provider,
+                    threshold_percent=self.context_compressor.threshold_percent,
                 )
 
             self._emit_status(
@@ -7690,6 +7693,7 @@ def _restore_primary_runtime(self) -> bool:
                 base_url=rt["compressor_base_url"],
                 api_key=rt["compressor_api_key"],
                 provider=rt["compressor_provider"],
+                threshold_percent=rt.get("compressor_threshold_percent"),
             )
 
             # ── Reset fallback chain for the new turn ──

tests/agent/test_update_model_threshold_percent.py

@@ -0,0 +1,71 @@
+"""Tests for ContextCompressor.update_model() threshold_percent handling.
+
+Regression test for #18617: threshold_percent not updated on model switch.
+"""
+import pytest
+from unittest.mock import patch
+
+
+def _make_compressor(model="model-a", threshold_percent=0.70, context_length=200000):
+    """Create a ContextCompressor with controlled context length."""
+    from agent.context_compressor import ContextCompressor
+    with patch("agent.context_compressor.get_model_context_length", return_value=context_length):
+        return ContextCompressor(
+            model=model,
+            threshold_percent=threshold_percent,
+            quiet_mode=True,
+        )
+
+
+class TestUpdateModelThresholdPercent:
+    """update_model() should accept and apply threshold_percent."""
+
+    def test_update_model_preserves_threshold_percent_when_not_passed(self):
+        """When threshold_percent is not passed, the existing value is kept."""
+        cc = _make_compressor(threshold_percent=0.70)
+        assert cc.threshold_percent == 0.70
+
+        cc.update_model(model="model-b", context_length=100000)
+        # threshold_percent unchanged
+        assert cc.threshold_percent == 0.70
+        # threshold_tokens recalculated from new context_length
+        # 100000 * 0.70 = 70000 > MINIMUM_CONTEXT_LENGTH (64000)
+        assert cc.threshold_tokens == 70000
+
+    def test_update_model_accepts_threshold_percent_parameter(self):
+        """update_model() should accept threshold_percent and update it."""
+        cc = _make_compressor(threshold_percent=0.70)
+        cc.update_model(model="model-b", context_length=100000, threshold_percent=0.50)
+        assert cc.threshold_percent == 0.50
+
+    def test_update_model_recalculates_threshold_tokens_with_new_percent(self):
+        """After threshold_percent update, threshold_tokens uses the new value."""
+        cc = _make_compressor(threshold_percent=0.70, context_length=200000)
+        assert cc.threshold_percent == 0.70
+        # 200000 * 0.70 = 140000
+        assert cc.threshold_tokens == 140000
+
+        cc.update_model(model="model-b", context_length=100000, threshold_percent=0.80)
+        # 100000 * 0.80 = 80000 > MINIMUM_CONTEXT_LENGTH
+        assert cc.threshold_percent == 0.80
+        assert cc.threshold_tokens == 80000
+
+    def test_update_model_preserves_percent_on_context_change_only(self):
+        """Model switch with same percent should still recalculate tokens."""
+        cc = _make_compressor(threshold_percent=0.60, context_length=200000)
+        assert cc.threshold_tokens == 120000
+
+        cc.update_model(model="model-b", context_length=300000)
+        assert cc.threshold_percent == 0.60
+        # 300000 * 0.60 = 180000
+        assert cc.threshold_tokens == 180000
+
+    def test_threshold_floor_still_applies(self):
+        """MINIMUM_CONTEXT_LENGTH floor should still apply after percent update."""
+        cc = _make_compressor(threshold_percent=0.70, context_length=200000)
+        # Switch to small model with low percent that would be below floor
+        cc.update_model(model="small-model", context_length=80000, threshold_percent=0.50)
+        # 80000 * 0.50 = 40000 < 64000 (MINIMUM_CONTEXT_LENGTH)
+        # Floor should kick in
+        from agent.model_metadata import MINIMUM_CONTEXT_LENGTH
+        assert cc.threshold_tokens == MINIMUM_CONTEXT_LENGTH