@miller-tech/uap 1.195.1 → 1.195.2

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@miller-tech/uap",
-  "version": "1.195.1",
+  "version": "1.195.2",
   "description": "Autonomous AI agent memory system with CLAUDE.md protocol enforcement",
   "type": "module",
   "main": "dist/index.js",
@@ -21,7 +21,7 @@
     "start": "node dist/bin/cli.js",
     "test": "vitest",
     "test:ci": "vitest run",
-    "test:enforcers": "UAP_PROXY_ENV_AUTOLOAD=0 python3 -m unittest tools.agents.tests.test_enforcement_self_protect tools.agents.tests.test_schema_diff_gate tools.agents.tests.test_stream_telemetry tools.agents.tests.test_project_telemetry_events tools.agents.tests.test_workdir_scope_enforcer tools.agents.tests.test_gate_evidence tools.agents.tests.test_repeat_call_guard tools.agents.tests.test_gate_integrity tools.agents.tests.test_gate_failclosed_indirection tools.agents.tests.test_expert_review_pr_scope tools.agents.tests.test_path_containment tools.agents.tests.test_path_normalizer_hardened tools.agents.tests.test_sandbox_tool_strip tools.agents.tests.test_proxy_env_loader tools.agents.tests.test_stream_required_tool tools.agents.tests.test_tool_call_wrapper_profiles tools.agents.tests.test_tool_convert_cache tools.agents.tests.test_doubling_break tools.agents.tests.test_error_loop_ignores_correctives tools.agents.tests.test_attractor_detection tools.agents.tests.test_client_disconnect tools.agents.tests.test_confidence_escalation tools.agents.tests.test_coordination_ban tools.agents.tests.test_coordination_early_ban tools.agents.tests.test_cycle_break_exploration tools.agents.tests.test_deferral_break tools.agents.tests.test_deliver_autoroute tools.agents.tests.test_delivery_enforcement_all_langs tools.agents.tests.test_delivery_enforcement_exemptions tools.agents.tests.test_delivery_enforcement_filepath tools.agents.tests.test_delivery_enforcement_write_complexity tools.agents.tests.test_delivery_enforcement_web_and_bash tools.agents.tests.test_disconnect_watcher tools.agents.tests.test_empty_maxtokens_recovery tools.agents.tests.test_empty_tool_loop_break tools.agents.tests.test_enforcer_escape_hatches tools.agents.tests.test_error_loop_break tools.agents.tests.test_finalize_suppression tools.agents.tests.test_malformed_unclosed_think tools.agents.tests.test_mandate_beats_recon tools.agents.tests.test_mandate_deliver tools.agents.tests.test_overflow_truncate_count_tokens tools.agents.tests.test_passthrough_oauth tools.agents.tests.test_project_telemetry tools.agents.tests.test_proxy_auth_headers tools.agents.tests.test_prune_preserve_force_write tools.agents.tests.test_recon_deliver_gate tools.agents.tests.test_session_admission tools.agents.tests.test_stream_heartbeat tools.agents.tests.test_stuck_break_reattach tools.agents.tests.test_turn_count_breaker_periodic tools.agents.tests.test_upstream_chokepoint tools.agents.tests.test_vision_passthrough tools.agents.tests.test_worktree_required tools.agents.tests.test_enforcer_suite_coverage tools.agents.tests.test_validate_plan_gate tools.agents.tests.test_validate_plan_inside_project tools.agents.tests.test_anthropic_proxy_streaming tools.agents.tests.test_delivery_enforcement_worktree tools.agents.tests.test_output_token_ceilings tools.agents.tests.test_tool_narrowing_core tools.agents.tests.test_models_context_window",
+    "test:enforcers": "UAP_PROXY_ENV_AUTOLOAD=0 python3 -m unittest tools.agents.tests.test_enforcement_self_protect tools.agents.tests.test_schema_diff_gate tools.agents.tests.test_stream_telemetry tools.agents.tests.test_project_telemetry_events tools.agents.tests.test_workdir_scope_enforcer tools.agents.tests.test_gate_evidence tools.agents.tests.test_repeat_call_guard tools.agents.tests.test_gate_integrity tools.agents.tests.test_gate_failclosed_indirection tools.agents.tests.test_expert_review_pr_scope tools.agents.tests.test_path_containment tools.agents.tests.test_path_normalizer_hardened tools.agents.tests.test_sandbox_tool_strip tools.agents.tests.test_proxy_env_loader tools.agents.tests.test_stream_required_tool tools.agents.tests.test_tool_call_wrapper_profiles tools.agents.tests.test_tool_convert_cache tools.agents.tests.test_doubling_break tools.agents.tests.test_error_loop_ignores_correctives tools.agents.tests.test_attractor_detection tools.agents.tests.test_client_disconnect tools.agents.tests.test_confidence_escalation tools.agents.tests.test_coordination_ban tools.agents.tests.test_coordination_early_ban tools.agents.tests.test_cycle_break_exploration tools.agents.tests.test_deferral_break tools.agents.tests.test_deliver_autoroute tools.agents.tests.test_delivery_enforcement_all_langs tools.agents.tests.test_delivery_enforcement_exemptions tools.agents.tests.test_delivery_enforcement_filepath tools.agents.tests.test_delivery_enforcement_write_complexity tools.agents.tests.test_delivery_enforcement_web_and_bash tools.agents.tests.test_disconnect_watcher tools.agents.tests.test_empty_maxtokens_recovery tools.agents.tests.test_empty_tool_loop_break tools.agents.tests.test_enforcer_escape_hatches tools.agents.tests.test_error_loop_break tools.agents.tests.test_finalize_suppression tools.agents.tests.test_malformed_unclosed_think tools.agents.tests.test_mandate_beats_recon tools.agents.tests.test_mandate_deliver tools.agents.tests.test_overflow_truncate_count_tokens tools.agents.tests.test_passthrough_oauth tools.agents.tests.test_project_telemetry tools.agents.tests.test_proxy_auth_headers tools.agents.tests.test_prune_preserve_force_write tools.agents.tests.test_recon_deliver_gate tools.agents.tests.test_session_admission tools.agents.tests.test_stream_heartbeat tools.agents.tests.test_stuck_break_reattach tools.agents.tests.test_turn_count_breaker_periodic tools.agents.tests.test_upstream_chokepoint tools.agents.tests.test_vision_passthrough tools.agents.tests.test_worktree_required tools.agents.tests.test_enforcer_suite_coverage tools.agents.tests.test_validate_plan_gate tools.agents.tests.test_validate_plan_inside_project tools.agents.tests.test_anthropic_proxy_streaming tools.agents.tests.test_non_stream_resp_log tools.agents.tests.test_delivery_enforcement_worktree tools.agents.tests.test_output_token_ceilings tools.agents.tests.test_tool_narrowing_core tools.agents.tests.test_models_context_window",
     "test:coverage": "vitest --coverage",
     "bench": "vitest --config vitest.bench.config.ts",
     "lint": "eslint src --ext .ts",

package/tools/agents/scripts/anthropic_proxy.py

@@ -10504,6 +10504,43 @@ def _extract_thinking_block(text: str) -> tuple[str | None, str]:
     return "\n\n".join(p for p in parts if p), remaining.lstrip()
 
 
+def _log_non_stream_resp(content: list, finish: str, usage: dict) -> None:
+    """Log what a NON-STREAMING turn produced, in the streaming line's shape.
+
+    ``RESP:`` was emitted only from ``stream_anthropic_response``, so every
+    non-streaming turn logged a request and no outcome. That is not a small
+    gap: ``uap deliver``'s agentic executor is entirely non-streaming, so the
+    journal recorded 431 requests and ONE response during a three-hour run on
+    2026-08-11 — and diagnosing that run meant reconstructing what the model
+    had done from the delivery log and the tool counts instead of reading it.
+    A monitor that cannot see the answers cannot tell working from looping.
+
+    Same field order and prefix as the streaming line ON PURPOSE, so existing
+    log analysis picks both up without a second parser; ``path=json`` is
+    appended for anyone who needs to tell them apart.
+    """
+    text = "".join(
+        b.get("text", "") for b in content if isinstance(b, dict) and b.get("type") == "text"
+    )
+    tool_names = [
+        b.get("name") for b in content if isinstance(b, dict) and b.get("type") == "tool_use"
+    ]
+    tool_args = [
+        json.dumps(b.get("input", {}), separators=(",", ":"))[:200]
+        for b in content
+        if isinstance(b, dict) and b.get("type") == "tool_use"
+    ]
+    logger.info(
+        "RESP: finish=%s output_tokens=%d text_len=%d text=%.300s tool_calls=%s args=%s path=json",
+        finish,
+        usage.get("completion_tokens", 0),
+        len(text),
+        text[:300],
+        tool_names,
+        tool_args,
+    )
+
+
 def openai_to_anthropic_response(
     openai_resp: dict,
     model: str,
@@ -10646,6 +10683,8 @@ def openai_to_anthropic_response(
 
     usage = openai_resp.get("usage", {})
 
+    _log_non_stream_resp(content, finish, usage)
+
     return {
         "id": f"msg_{uuid.uuid4().hex[:24]}",
         "type": "message",

package/tools/agents/tests/test_non_stream_resp_log.py

@@ -0,0 +1,117 @@
+#!/usr/bin/env python3
+"""A monitor that cannot see the answers cannot tell working from looping.
+
+`RESP:` was logged only from `stream_anthropic_response`. `uap deliver`'s
+agentic executor is entirely non-streaming, so during a three-hour run on
+2026-08-11 the journal held 431 requests and ONE response — and every one of
+those 431 turns was invisible. Diagnosing that run meant reconstructing the
+model's behaviour from the delivery log's tool counts instead of reading what
+it actually returned.
+
+These pin that a non-streaming turn now logs its outcome, in the same shape as
+the streaming one so a single parser reads both.
+"""
+
+import importlib.util
+import json
+import logging
+import unittest
+from pathlib import Path
+
+
+def _load_proxy_module():
+    proxy_path = Path(__file__).resolve().parents[1] / "scripts" / "anthropic_proxy.py"
+    spec = importlib.util.spec_from_file_location("anthropic_proxy", proxy_path)
+    assert spec is not None and spec.loader is not None
+    module = importlib.util.module_from_spec(spec)
+    spec.loader.exec_module(module)
+    return module
+
+
+proxy = _load_proxy_module()
+
+
+def _resp(message, finish="stop", usage=None):
+    return {
+        "choices": [{"message": message, "finish_reason": finish}],
+        "usage": usage or {"prompt_tokens": 10, "completion_tokens": 7},
+    }
+
+
+class NonStreamRespLog(unittest.TestCase):
+    def _convert(self, openai_resp):
+        with self.assertLogs("uap.anthropic_proxy", level="INFO") as captured:
+            out = proxy.openai_to_anthropic_response(openai_resp, "qwen")
+        return out, [r for r in captured.output if "RESP: finish=" in r]
+
+    def test_text_turn_logs_its_outcome(self):
+        _, lines = self._convert(_resp({"role": "assistant", "content": "hello world"}))
+        self.assertEqual(len(lines), 1, "a non-streaming turn must log exactly one RESP")
+        line = lines[0]
+        self.assertIn("finish=stop", line)
+        self.assertIn("output_tokens=7", line)
+        self.assertIn("text_len=11", line)
+        self.assertIn("hello world", line)
+
+    def test_tool_turn_names_the_tool_and_its_args(self):
+        # The whole point: "what did it DO this round" has to be readable.
+        _, lines = self._convert(
+            _resp(
+                {
+                    "role": "assistant",
+                    "content": None,
+                    "tool_calls": [
+                        {
+                            "id": "call_1",
+                            "type": "function",
+                            "function": {
+                                "name": "read_file",
+                                "arguments": json.dumps({"path": "setup.sql", "offset": 607}),
+                            },
+                        }
+                    ],
+                },
+                finish="tool_calls",
+            )
+        )
+        self.assertEqual(len(lines), 1)
+        self.assertIn("read_file", lines[0])
+        self.assertIn("setup.sql", lines[0])
+        self.assertIn("607", lines[0], "the args are what distinguish paging from a re-read")
+
+    def test_empty_completion_is_visible_as_empty(self):
+        # An empty completion is a failure mode with its own history here
+        # (decode-compliance/budget truncation). It must not read as silence.
+        _, lines = self._convert(
+            _resp({"role": "assistant", "content": ""}, finish="length",
+                  usage={"prompt_tokens": 5, "completion_tokens": 0})
+        )
+        self.assertEqual(len(lines), 1)
+        self.assertIn("text_len=0", lines[0])
+        self.assertIn("finish=length", lines[0])
+
+    def test_shape_matches_the_streaming_line_so_one_parser_reads_both(self):
+        _, lines = self._convert(_resp({"role": "assistant", "content": "x"}))
+        for field in ("RESP: finish=", "output_tokens=", "text_len=", "text=", "tool_calls=", "args="):
+            self.assertIn(field, lines[0], field)
+        self.assertIn("path=json", lines[0], "…while still being distinguishable")
+
+    def test_logging_never_breaks_the_conversion(self):
+        # Fail-soft: the response is the product, the log line is not. A
+        # malformed content block must not cost the client its answer.
+        out, _ = self._convert(_resp({"role": "assistant", "content": "fine"}))
+        self.assertEqual(out["role"], "assistant")
+        self.assertEqual(out["content"][0]["text"], "fine")
+        self.assertEqual(out["stop_reason"], "end_turn")
+
+
+class RespLogSurvivesOddContent(unittest.TestCase):
+    def test_non_dict_content_blocks_do_not_raise(self):
+        logging.getLogger("uap.anthropic_proxy").setLevel(logging.INFO)
+        # Defensive: content is assembled upstream and has been non-uniform
+        # before (thinking promotion, text-tool extraction).
+        proxy._log_non_stream_resp(["not a dict", {"type": "text", "text": "ok"}], "stop", {})
+
+
+if __name__ == "__main__":
+    unittest.main()