UiPath · tmatup · Sep 28, 2026 · Sep 28, 2026
@@ -210,7 +210,10 @@ token/cost budget breach, which fires only after every criterion is already scor
 cache buckets, captured proxy-side on the LiteLLM open-weight backend and rendered
 by the evalboard as a per-call table; empty on every other
 backend), `num_turns`, `max_turns_exhausted`,
-`result_summary` (`{is_error, subtype, stop_reason, result}`), `crashed`,
+`result_summary` (`{is_error, subtype, stop_reason, result}`), `provider_errors`
+(`list[ProviderError]` — `{at, message, kind, http_status, will_retry, details}`, one
+row per model-provider error the agent's own client reported, including errors it
+retried itself; filled by the codex adapter, empty on other backends), `crashed`,
 `crash_reason`.
 
 > **Token invariant.** Summing the four token buckets across `messages`

@@ -11,6 +11,7 @@
 import time
 from collections.abc import Callable
 from datetime import datetime
+from enum import Enum
 from pathlib import Path
 from typing import Any, ClassVar, NamedTuple
 from urllib.parse import urlparse
@@ -33,6 +34,7 @@
     CommandTelemetry,
     ContentBlock,
     DirectRoute,
+    ProviderError,
     SystemPromptSemantics,
     TokenUsage,
     TranscriptMessage,
@@ -274,6 +276,34 @@ def _message_uncached_input(m: AssistantMessage) -> int:
 _ZSH_PROFILE_NAMES = frozenset({".zshenv", ".zprofile", ".zshrc"})
 
 
+def _provider_error_from(payload: Any) -> ProviderError:
+    """Map a codex ``ErrorNotification`` payload onto a ``ProviderError``.
+
+    ``codexErrorInfo`` is either a bare category (enum) or a one-key object whose
+    key is the category and whose value may carry ``httpStatusCode``.
+    """
+    error = getattr(payload, "error", None)
+    kind: str | None = None
+    http_status: int | None = None
+    info = getattr(getattr(error, "codex_error_info", None), "root", None)
+    if isinstance(info, Enum):
+        kind = str(info.value)
+    elif hasattr(info, "model_dump"):
+        dumped = info.model_dump(by_alias=True, mode="json")
+        if len(dumped) == 1:
+            kind, body = next(iter(dumped.items()))
+            if isinstance(body, dict) and isinstance(body.get("httpStatusCode"), int):
+                http_status = body["httpStatusCode"]
+    return ProviderError(
+        at=datetime.now(),
+        message=str(getattr(error, "message", "") or ""),
+        kind=kind,
+        http_status=http_status,
+        will_retry=bool(getattr(payload, "will_retry", False)),
+        details=getattr(error, "additional_details", None),
+    )
+
+
 def _get_item_root(notification: Any) -> Any:
     """Extract the typed item root from a Codex SDK notification.
 
@@ -338,6 +368,7 @@ def __init__(
         # Live pump scratch (set during streaming).
         self.turn_result: Any = None
         self.latest_token_usage: Any = None
+        self.provider_errors: list[ProviderError] = []
         self.agent_message_chunks: list[str] = []
         # Sequence per executable item, assigned at item/started and reused at
         # item/completed via this id->seq map.
@@ -546,8 +577,19 @@ def dispatch(self, notification: Any) -> bool:
             self.on_token_usage_updated(notification)
         elif method == "turn/completed":
             return self.on_turn_completed(notification)
+        elif method == "error":
+            self.on_error(notification)
         return False
 
+    def on_error(self, notification: Any) -> None:
+        """Record a provider error; codex retries a stalled stream silently otherwise."""
+        err = _provider_error_from(notification.payload)
+        self.provider_errors.append(err)
+        self._agent._log.warning(
+            f"provider error (will_retry={err.will_retry}, kind={err.kind}, "
+            + f"http_status={err.http_status}): {err.message}"
+        )
+
     def on_item_started(self, notification: Any) -> None:
         """Emit ToolStartEvent + record the tool_use block for every tool-like item."""
         root = _get_item_root(notification)
@@ -765,6 +807,7 @@ def finalize(self, status: AgentEndStatus, *, crashed: bool = False, crash_reaso
                 assistant_turn_count=1,
                 messages=self.messages,
                 num_turns=max(self.api_calls, 1),
+                provider_errors=list(self.provider_errors),
                 crashed=crashed,
                 crash_reason=crash_reason,
                 max_turns_exhausted=status is AgentEndStatus.MAX_TURNS_EXHAUSTED,

@@ -228,6 +228,7 @@
     CommandTelemetry,
     ContentBlock,
     ProviderCallCost,
+    ProviderError,
     ReconciliationMessage,
     SlowestCommandInfo,
     TokenUsage,
@@ -349,6 +350,7 @@
     "CommandStatistics",
     "ContentBlock",
     "ProviderCallCost",
+    "ProviderError",
     "ReconciliationMessage",
     "SlowestCommandInfo",
     "TokenUsage",

@@ -26,6 +26,7 @@
     CommandStatistics,
     CommandTelemetry,
     ProviderCallCost,
+    ProviderError,
     TokenUsage,
     TranscriptMessage,
 )
@@ -417,6 +418,14 @@ class TurnRecord(BaseModel):
             "these calls' cost_usd (the real bill), not the static rate-card estimate."
         ),
     )
+    provider_errors: list[ProviderError] = Field(
+        default_factory=list,
+        description=(
+            "Model-provider errors the agent's own client reported during this turn, in arrival "
+            "order, including ones it retried itself (will_retry=True). Filled by the codex "
+            "adapter from its `error` notification; empty on backends that do not surface them."
+        ),
+    )
     crashed: bool = Field(
         default=False,
         description=(

@@ -134,6 +134,24 @@ class ProviderCallCost(BaseModel):
     output_tokens: int | None = Field(default=None, description="Completion tokens for this call.")
 
 
+class ProviderError(BaseModel):
+    """One model-provider error the agent's own client reported mid-turn.
+
+    Codex retries a stalled or failed provider stream internally and says so only
+    through its ``error`` notification (``willRetry``). Without this record a
+    9-minute provider stall reads as pure model latency.
+    """
+
+    at: datetime = Field(description="When the harness received the error notification.")
+    message: str = Field(description="The client's error message.")
+    kind: str | None = Field(
+        default=None, description="Client error category (codex `codexErrorInfo`), e.g. responseStreamDisconnected."
+    )
+    http_status: int | None = Field(default=None, description="Upstream HTTP status, when the client forwarded one.")
+    will_retry: bool = Field(description="True when the client retries the request itself after this error.")
+    details: str | None = Field(default=None, description="Extra detail the client attached, if any.")
+
+
 class ContentBlock(BaseModel):
     """One content block within a message, in emission order.
 

@@ -224,6 +224,7 @@ def build_turn_record(self) -> TurnRecord:
             num_turns=end.num_turns,
             max_turns_exhausted=end.max_turns_exhausted,
             result_summary=end.result_summary,
+            provider_errors=end.provider_errors,
             crashed=end.crashed,
             crash_reason=end.crash_reason,
             harness_startup_ms=startup_ms,

@@ -32,6 +32,7 @@
 
 from coder_eval.models import (
     CommandTelemetry,
+    ProviderError,
     ResultSummary,
     TokenUsage,
     TranscriptMessage,
@@ -169,6 +170,7 @@ class AgentEndEvent(StreamEvent):
     num_turns: int | None = None
     max_turns_exhausted: bool = False
     result_summary: ResultSummary | None = None
+    provider_errors: list[ProviderError] = Field(default_factory=list)
     crashed: bool = False
     crash_reason: str | None = None
     duration_seconds: float = 0.0

@@ -40,6 +40,7 @@
   ],
   "model_used": "gemini-3.5-flash",
   "num_turns": 1,
+  "provider_errors": [],
   "result_summary": null,
   "timestamp": "<scrubbed>",
   "token_usage": {

@@ -71,6 +71,7 @@
   ],
   "model_used": "gemini-3.5-flash",
   "num_turns": 1,
+  "provider_errors": [],
   "result_summary": null,
   "timestamp": "<scrubbed>",
   "token_usage": {

@@ -80,6 +80,7 @@
   ],
   "model_used": "gemini-3.5-flash",
   "num_turns": 1,
+  "provider_errors": [],
   "result_summary": null,
   "timestamp": "<scrubbed>",
   "token_usage": {

@@ -60,6 +60,7 @@
   ],
   "model_used": "gemini-3.5-flash",
   "num_turns": 1,
+  "provider_errors": [],
   "result_summary": null,
   "timestamp": "<scrubbed>",
   "token_usage": {

@@ -94,6 +94,7 @@
   ],
   "model_used": "gemini-3.5-flash",
   "num_turns": 3,
+  "provider_errors": [],
   "result_summary": null,
   "timestamp": "<scrubbed>",
   "token_usage": {

@@ -40,6 +40,7 @@
   ],
   "model_used": "mock-model",
   "num_turns": 1,
+  "provider_errors": [],
   "result_summary": {
     "is_error": false,
     "result": "done",

@@ -62,6 +62,7 @@
   ],
   "model_used": "mock-model",
   "num_turns": 1,
+  "provider_errors": [],
   "result_summary": {
     "is_error": false,
     "result": "done",

@@ -116,6 +116,7 @@
   ],
   "model_used": "mock-model",
   "num_turns": 1,
+  "provider_errors": [],
   "result_summary": {
     "is_error": false,
     "result": "done",

@@ -89,6 +89,7 @@
   ],
   "model_used": "mock-model",
   "num_turns": 1,
+  "provider_errors": [],
   "result_summary": {
     "is_error": false,
     "result": "done",

@@ -48,6 +48,7 @@
   ],
   "model_used": "mock-model",
   "num_turns": 1,
+  "provider_errors": [],
   "result_summary": {
     "is_error": false,
     "result": "done",

@@ -63,6 +63,7 @@
   ],
   "model_used": "mock-model",
   "num_turns": 1,
+  "provider_errors": [],
   "result_summary": {
     "is_error": false,
     "result": "done",

@@ -12,6 +12,7 @@
   "messages": [],
   "model_used": null,
   "num_turns": null,
+  "provider_errors": [],
   "result_summary": null,
   "timestamp": "<scrubbed>",
   "token_usage": null,

@@ -12,6 +12,7 @@
   "messages": [],
   "model_used": null,
   "num_turns": null,
+  "provider_errors": [],
   "result_summary": null,
   "timestamp": "<scrubbed>",
   "token_usage": null,

@@ -12,6 +12,7 @@
   "messages": [],
   "model_used": null,
   "num_turns": null,
+  "provider_errors": [],
   "result_summary": null,
   "timestamp": "<scrubbed>",
   "token_usage": null,

@@ -40,6 +40,7 @@
   ],
   "model_used": "mock-model",
   "num_turns": null,
+  "provider_errors": [],
   "result_summary": null,
   "timestamp": "<scrubbed>",
   "token_usage": null,

@@ -40,6 +40,7 @@
   ],
   "model_used": "gpt-5-codex",
   "num_turns": 1,
+  "provider_errors": [],
   "result_summary": null,
   "timestamp": "<scrubbed>",
   "token_usage": {

@@ -71,6 +71,7 @@
   ],
   "model_used": "gpt-5-codex",
   "num_turns": 2,
+  "provider_errors": [],
   "result_summary": null,
   "timestamp": "<scrubbed>",
   "token_usage": {

@@ -67,6 +67,7 @@
   ],
   "model_used": "gpt-5-codex",
   "num_turns": 1,
+  "provider_errors": [],
   "result_summary": null,
   "timestamp": "<scrubbed>",
   "token_usage": {

@@ -62,6 +62,7 @@
   ],
   "model_used": "gpt-5-codex",
   "num_turns": 2,
+  "provider_errors": [],
   "result_summary": null,
   "timestamp": "<scrubbed>",
   "token_usage": {

@@ -62,6 +62,7 @@
   ],
   "model_used": "gpt-5-codex",
   "num_turns": 1,
+  "provider_errors": [],
   "result_summary": null,
   "timestamp": "<scrubbed>",
   "token_usage": null,

@@ -120,6 +120,7 @@
   ],
   "model_used": "gpt-5-codex",
   "num_turns": 1,
+  "provider_errors": [],
   "result_summary": null,
   "timestamp": "<scrubbed>",
   "token_usage": null,

@@ -40,6 +40,7 @@
   ],
   "model_used": "gpt-5-codex",
   "num_turns": 1,
+  "provider_errors": [],
   "result_summary": null,
   "timestamp": "<scrubbed>",
   "token_usage": null,

@@ -40,6 +40,7 @@
   ],
   "model_used": "gpt-5-codex",
   "num_turns": 1,
+  "provider_errors": [],
   "result_summary": null,
   "timestamp": "<scrubbed>",
   "token_usage": {

@@ -40,6 +40,7 @@
   ],
   "model_used": "deepseek/deepseek-v4-pro",
   "num_turns": 1,
+  "provider_errors": [],
   "result_summary": {
     "is_error": false,
     "result": null,

@@ -89,6 +89,7 @@
   ],
   "model_used": "deepseek/deepseek-v4-pro",
   "num_turns": 2,
+  "provider_errors": [],
   "result_summary": {
     "is_error": false,
     "result": null,

@@ -89,6 +89,7 @@
   ],
   "model_used": "deepseek/deepseek-v4-pro",
   "num_turns": 2,
+  "provider_errors": [],
   "result_summary": {
     "is_error": false,
     "result": null,