@miller-tech/uap 1.20.9 → 1.20.10

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@miller-tech/uap",
-  "version": "1.20.9",
+  "version": "1.20.10",
   "description": "Autonomous AI agent memory system with CLAUDE.md protocol enforcement",
   "type": "module",
   "main": "dist/index.js",

package/tools/agents/scripts/anthropic_proxy.py

@@ -280,9 +280,6 @@ PROXY_FORCED_TOOL_DAMPENER_REJECTIONS = int(
 PROXY_TOOL_STARVATION_THRESHOLD = int(
     os.environ.get("PROXY_TOOL_STARVATION_THRESHOLD", "5")
 )
-PROXY_CONTEXT_HIGH_RELAXATION_THRESHOLD = float(
-    os.environ.get("PROXY_CONTEXT_HIGH_RELAXATION_THRESHOLD", "0.70")
-)
 PROXY_SESSION_CONTAMINATION_BREAKER = os.environ.get(
     "PROXY_SESSION_CONTAMINATION_BREAKER", "on"
 ).lower() not in {
@@ -2388,22 +2385,6 @@ def build_openai_request(
             if not has_tool_results:
                 monitor.reset_tool_turn_state(reason="no_tool_results")
 
-        # CONTEXT-AWARE RELAXATION: when context utilization is high and
-        # tool_choice was forced to required, relax to auto to let the model
-        # emit shorter text responses instead of consuming more tokens.
-        if openai_body.get("tool_choice") == "required":
-            ctx_utilization = (
-                monitor.last_input_tokens / monitor.context_window
-                if monitor.context_window > 0
-                else 0.0
-            )
-            if ctx_utilization >= PROXY_CONTEXT_HIGH_RELAXATION_THRESHOLD:
-                openai_body["tool_choice"] = "auto"
-                logger.warning(
-                    "CONTEXT-AWARE RELAXATION: tool_choice=auto (utilization=%.1f%% >= %.0f%% threshold)",
-                    ctx_utilization * 100,
-                    PROXY_CONTEXT_HIGH_RELAXATION_THRESHOLD * 100,
-                )
 
         if PROXY_DISABLE_THINKING_ON_TOOL_TURNS:
             openai_body["enable_thinking"] = False

package/tools/agents/tests/test_anthropic_proxy_streaming.py

@@ -3166,44 +3166,6 @@ class TestToolStarvationBreaker(unittest.TestCase):
         self.assertIn("tools", result)
 
 
-class TestContextAwareRelaxation(unittest.TestCase):
-    """Tests for context-aware tool_choice relaxation."""
-
-    def test_relaxes_at_high_utilization(self):
-        monitor = proxy.SessionMonitor()
-        monitor.context_window = 100000
-        monitor.last_input_tokens = 75000  # 75% > 70% threshold
-        body = {
-            "model": "qwen3.5",
-            "messages": [
-                {"role": "user", "content": "hello"},
-                {"role": "assistant", "content": "text only"},
-                {"role": "user", "content": [{"type": "tool_result", "tool_use_id": "x", "content": "ok"}]},
-            ],
-            "tools": [{"name": "Bash", "input_schema": {"type": "object", "properties": {"command": {"type": "string"}}}}],
-        }
-        result = proxy.build_openai_request(body, monitor)
-        # Should be auto, not required
-        self.assertEqual(result.get("tool_choice"), "auto")
-
-    def test_no_relaxation_below_threshold(self):
-        monitor = proxy.SessionMonitor()
-        monitor.context_window = 100000
-        monitor.last_input_tokens = 50000  # 50% < 70%
-        body = {
-            "model": "qwen3.5",
-            "messages": [
-                {"role": "user", "content": "hello"},
-                {"role": "assistant", "content": "text only"},
-                {"role": "user", "content": [{"type": "tool_result", "tool_use_id": "x", "content": "ok"}]},
-            ],
-            "tools": [{"name": "Bash", "input_schema": {"type": "object", "properties": {"command": {"type": "string"}}}}],
-        }
-        result = proxy.build_openai_request(body, monitor)
-        # Should still be required (state machine forces it)
-        self.assertEqual(result.get("tool_choice"), "required")
-
-
 if __name__ == "__main__":
     unittest.main()