fix(core): preserve provider session failures

packages/llm/src/protocols/openai-responses.ts

@@ -891,6 +891,8 @@ const providerError = (event: OpenAIResponsesEvent, fallback: string) => {
   return LLMEvent.providerError({
     message,
     classification: code === "context_length_exceeded" || isContextOverflow(message) ? "context-overflow" : undefined,
+    category: code === "rate_limit_exceeded" ? "rate-limit" : undefined,
+    retryable: code === "rate_limit_exceeded" ? true : undefined,
   })
 }
 

packages/llm/src/schema/errors.ts

@@ -4,6 +4,20 @@ import { ModelID, ProviderID, ProviderMetadata, RouteID } from "./ids"
 export const ProviderFailureClassification = Schema.Literal("context-overflow")
 export type ProviderFailureClassification = typeof ProviderFailureClassification.Type
 
+export const ProviderFailureCategory = Schema.Literals([
+  "invalid-request",
+  "no-route",
+  "authentication",
+  "rate-limit",
+  "quota-exceeded",
+  "content-policy",
+  "provider-internal",
+  "transport",
+  "invalid-provider-output",
+  "unknown",
+])
+export type ProviderFailureCategory = typeof ProviderFailureCategory.Type
+
 export class HttpRequestDetails extends Schema.Class<HttpRequestDetails>("LLM.HttpRequestDetails")({
   method: Schema.String,
   url: Schema.String,

packages/llm/src/schema/events.ts

@@ -2,7 +2,7 @@ import { Schema } from "effect"
 import { ContentBlockID, FinishReason, ProtocolID, ProviderMetadata, RouteID, ToolCallID } from "./ids"
 import { ModelSchema } from "./options"
 import { ToolOutput, ToolResultValue } from "./messages"
-import { ProviderFailureClassification } from "./errors"
+import { ProviderFailureCategory, ProviderFailureClassification } from "./errors"
 
 /**
  * Token usage reported by an LLM provider.
@@ -201,6 +201,8 @@ export const ProviderErrorEvent = Schema.Struct({
   type: Schema.tag("provider-error"),
   message: Schema.String,
   classification: Schema.optional(ProviderFailureClassification),
+  category: Schema.optional(ProviderFailureCategory),
+  status: Schema.optional(Schema.Number),
   retryable: Schema.optional(Schema.Boolean),
   providerMetadata: Schema.optional(ProviderMetadata),
 }).annotate({ identifier: "LLM.Event.ProviderError" })

packages/llm/test/provider/openai-responses.test.ts

@@ -1326,7 +1326,14 @@ describe("OpenAI Responses route", () => {
       // sometimes-generic provider message. The bare message alone meant
       // production errors like rate limits were indistinguishable from
       // unrelated stream failures.
-      expect(response.events).toEqual([{ type: "provider-error", message: "rate_limit_exceeded: Slow down" }])
+      expect(response.events).toEqual([
+        {
+          type: "provider-error",
+          message: "rate_limit_exceeded: Slow down",
+          category: "rate-limit",
+          retryable: true,
+        },
+      ])
     }),
   )