fix: give final acceptance a shared retry budget and diagnostics

artemis/llm/diagnostics.py

@@ -13,6 +13,8 @@ from uuid import uuid4
 from PIL import Image
 
 from artemis.data_engine.context_vars import CURRENT_NODE_NAME, CURRENT_TRACE_ID
+from artemis.llm.final_review import ACTIVE_FINAL_REVIEW, FinalReviewBudgetExceeded
+from artemis.llm.reliability import LLMExhaustedError
 from artemis.llm.usage import usage_from_message
 from artemis.llm.stream_guard import ToolArgumentOverflow
 
@@ -21,6 +23,8 @@ ACTIVE_ATTEMPT: ContextVar["RequestAttempt | None"] = ContextVar("llm_attempt",
 
 
 def error_text(error: BaseException, timeout: float | None = None) -> str:
+    if isinstance(error, FinalReviewBudgetExceeded):
+        return str(error)
     if isinstance(error, TimeoutError):
         return f"模型请求在 {timeout:g} 秒内未完成" if timeout else "模型请求超时"
     text = str(error).strip() or type(error).__name__
@@ -34,6 +38,14 @@ def error_text(error: BaseException, timeout: float | None = None) -> str:
     return text[:1000]
 
 
+def final_review_error_text(error: Exception) -> str:
+    if isinstance(error, FinalReviewBudgetExceeded):
+        return str(error)
+    if isinstance(error, LLMExhaustedError):
+        return f"验收模型请求重试已用尽（{error.failure.category.value}）：{error_text(error)}"
+    return f"{type(error).__name__}: {error_text(error)}"
+
+
 def input_images(args: tuple, kwargs: dict) -> list[dict]:
     """Inspect only inline image headers; never fetch URLs or retain image data."""
     images = []
@@ -76,6 +88,9 @@ class RequestAttempt:
         self.clock = time.monotonic()
         self.last_publish = self.clock
         self.tools: dict[int | str, dict] = {}
+        self.review = ACTIVE_FINAL_REVIEW.get()
+        if self.review:
+            self.review.request_started(request_id, attempt)
         self.data = {
             "images": [],
             **metadata,
@@ -102,6 +117,8 @@ class RequestAttempt:
         self.data["updated_at"] = time.time()
         self.data["elapsed_seconds"] = round(time.monotonic() - self.clock, 3)
         self.last_publish = time.monotonic()
+        if self.review:
+            self.data["final_review"] = self.review.snapshot()
         if not self.engine or not getattr(self.engine, "current_session_id", None):
             return
         try:
@@ -203,14 +220,28 @@ class RequestAttempt:
             if (usage := usage_from_message(response)) is not None:
                 self.data["usage"] = usage
         self.data.update(phase=phase, ended_at=time.time())
+        if self.review:
+            self.review.request_finished(phase, category)
         self.publish()
 
     def retry(self, delay):
         self.data.update(
             phase="retry_wait", retry_at=time.time() + delay, retry_delay_seconds=delay
         )
+        if self.review:
+            self.review.request_finished("retry_wait")
         self.publish()
 
+    def cancel(self, error: BaseException | None = None):
+        if self.review and self.review.expired and not isinstance(error, KeyboardInterrupt):
+            self.finish(
+                "failed",
+                error=FinalReviewBudgetExceeded(self.review.total_seconds),
+                category="final_review_budget_exhausted",
+            )
+        else:
+            self.finish("cancelled", error=error, category="cancelled")
+
 
 async def observe_request_headers(request):
     """Correlate a wire request and record only hashes of routing headers."""