diff --git a/packages/opencode/src/provider/error.ts b/packages/opencode/src/provider/error.ts
index 21149a2cf389..51ef6668d3bd 100644
--- a/packages/opencode/src/provider/error.ts
+++ b/packages/opencode/src/provider/error.ts
@@ -105,45 +105,68 @@ export function parseStreamError(input: unknown): ParsedStreamError | undefined
   if (!body) return
 
   const responseBody = JSON.stringify(body)
-  if (body.type !== "error") return
-
-  switch (body?.error?.code) {
-    case "context_length_exceeded":
-      return {
-        type: "context_overflow",
-        message: "Input exceeds context window of this model",
-        responseBody,
-      }
-    case "insufficient_quota":
-      return {
-        type: "api_error",
-        message: "Quota exceeded. Check your plan and billing details.",
-        isRetryable: false,
-        responseBody,
-      }
-    case "usage_not_included":
-      return {
-        type: "api_error",
-        message: "To use Codex with your ChatGPT plan, upgrade to Plus: https://chatgpt.com/explore/plus.",
-        isRetryable: false,
-        responseBody,
-      }
-    case "invalid_prompt":
-      return {
-        type: "api_error",
-        message: typeof body?.error?.message === "string" ? body?.error?.message : "Invalid prompt.",
-        isRetryable: false,
-        responseBody,
-      }
-    case "server_is_overloaded":
-    case "server_error":
-      return {
-        type: "api_error",
-        message: typeof body?.error?.message === "string" ? body?.error?.message : "Server error.",
-        isRetryable: true,
-        responseBody,
-      }
+  const outerError = typeof body.error === "object" && body.error !== null ? body.error : undefined
+  const wrappedError =
+    typeof outerError?.error === "object" && outerError.error !== null ? outerError.error : outerError
+  if (body.type !== "error" && !wrappedError) return
+
+  const code = typeof wrappedError?.code === "string" ? wrappedError.code : undefined
+  const type = typeof wrappedError?.type === "string" ? wrappedError.type : undefined
+  const message =
+    typeof wrappedError?.message === "string" && wrappedError.message.length > 0
+      ? wrappedError.message
+      : typeof body?.message === "string" && body.message.length > 0
+        ? body.message
+        : undefined
+
+  const fromCode = (value: string | undefined, isRetryable = false): ParsedStreamError | undefined => {
+    switch (value) {
+      case "context_length_exceeded":
+        return {
+          type: "context_overflow",
+          message: "Input exceeds context window of this model",
+          responseBody,
+        }
+      case "insufficient_quota":
+        return {
+          type: "api_error",
+          message: "Quota exceeded. Check your plan and billing details.",
+          isRetryable: false,
+          responseBody,
+        }
+      case "usage_not_included":
+        return {
+          type: "api_error",
+          message: "To use Codex with your ChatGPT plan, upgrade to Plus: https://chatgpt.com/explore/plus.",
+          isRetryable: false,
+          responseBody,
+        }
+      case "invalid_prompt":
+        return {
+          type: "api_error",
+          message: message ?? "Invalid prompt.",
+          isRetryable: false,
+          responseBody,
+        }
+      case "server_is_overloaded":
+      case "server_error":
+      case "service_unavailable_error":
+        return {
+          type: "api_error",
+          message: message ?? "Server error.",
+          isRetryable,
+          responseBody,
+        }
+    }
   }
+
+  const byCode = fromCode(code, true)
+  if (byCode) return byCode
+
+  const byType = fromCode(type, true)
+  if (byType) return byType
+
+  if (code === undefined && body.type === "error") return
 }
 
 export type ParsedAPICallError =
diff --git a/packages/opencode/src/session/message-v2.ts b/packages/opencode/src/session/message-v2.ts
index c57fcd53c96f..85651960ce88 100644
--- a/packages/opencode/src/session/message-v2.ts
+++ b/packages/opencode/src/session/message-v2.ts
@@ -685,7 +685,7 @@ export function fromError(
         },
         { cause: e },
       ).toObject()
-    case APICallError.isInstance(e):
+    case APICallError.isInstance(e): {
       const parsed = ProviderError.parseAPICallError({
         providerID: ctx.providerID,
         error: e,
@@ -711,8 +711,32 @@ export function fromError(
         },
         { cause: e },
       ).toObject()
-    case e instanceof Error:
+    }
+    case e instanceof Error: {
+      const parsed = ProviderError.parseStreamError(e)
+      if (parsed) {
+        if (parsed.type === "context_overflow") {
+          return new ContextOverflowError(
+            {
+              message: parsed.message,
+              responseBody: parsed.responseBody,
+            },
+            { cause: e },
+          ).toObject()
+        }
+        return new APIError(
+          {
+            message: parsed.message,
+            isRetryable: parsed.isRetryable,
+            responseBody: parsed.responseBody,
+          },
+          {
+            cause: e,
+          },
+        ).toObject()
+      }
       return new NamedError.Unknown({ message: errorMessage(e) }, { cause: e }).toObject()
+    }
     default:
       try {
         const parsed = ProviderError.parseStreamError(e)
diff --git a/packages/opencode/src/session/retry.ts b/packages/opencode/src/session/retry.ts
index 4139665bd2bd..6698dd1b6036 100644
--- a/packages/opencode/src/session/retry.ts
+++ b/packages/opencode/src/session/retry.ts
@@ -145,6 +145,12 @@ export function retryable(error: Err, provider: string) {
   if (code.includes("exhausted") || code.includes("unavailable")) {
     return { message: "Provider is overloaded" }
   }
+  if (
+    json.type === "error" &&
+    (json.error?.code === "server_is_overloaded" || json.error?.type === "service_unavailable_error")
+  ) {
+    return { message: typeof json.error?.message === "string" ? json.error.message : "Provider is overloaded" }
+  }
   if (json.type === "error" && typeof json.error?.code === "string" && json.error.code.includes("rate_limit")) {
     return { message: "Rate Limited" }
   }
diff --git a/packages/opencode/src/tool/task.ts b/packages/opencode/src/tool/task.ts
index 1ac9d4d9ad87..ffa82a48f1be 100644
--- a/packages/opencode/src/tool/task.ts
+++ b/packages/opencode/src/tool/task.ts
@@ -9,6 +9,7 @@ import { MessageV2 } from "../session/message-v2"
 import { Agent } from "../agent/agent"
 import { deriveSubagentSessionPermission } from "../agent/subagent-permissions"
 import type { SessionPrompt } from "../session/prompt"
+import { errorMessage } from "@/util/error"
 import { Config } from "@/config/config"
 import { Effect, Exit, Schema, Scope } from "effect"
 import { EffectBridge } from "@/effect/bridge"
@@ -188,6 +189,9 @@ export const TaskTool = Tool.define(
           },
           parts,
         })
+        if (result.info.role === "assistant" && result.info.error) {
+          return yield* Effect.fail(new Error(`Child assistant failed: ${errorMessage(result.info.error)}`))
+        }
         return result.parts.findLast((item) => item.type === "text")?.text ?? ""
       })
 
diff --git a/packages/opencode/test/session/retry.test.ts b/packages/opencode/test/session/retry.test.ts
index f5edf1af24b1..159e504530ea 100644
--- a/packages/opencode/test/session/retry.test.ts
+++ b/packages/opencode/test/session/retry.test.ts
@@ -127,6 +127,23 @@ describe("session.retry.retryable", () => {
     expect(SessionRetry.retryable(error, retryProvider)).toEqual({ message: "Provider is overloaded" })
   })
 
+  test("maps persisted nested OpenAI overload errors", () => {
+    const error = wrap(
+      JSON.stringify({
+        type: "error",
+        error: {
+          type: "service_unavailable_error",
+          code: "server_is_overloaded",
+          message: "Our servers are currently overloaded. Please try again later.",
+        },
+      }),
+    )
+
+    expect(SessionRetry.retryable(error, retryProvider)).toEqual({
+      message: "Our servers are currently overloaded. Please try again later.",
+    })
+  })
+
   test("does not retry unknown json messages", () => {
     const error = wrap(JSON.stringify({ error: { message: "no_kv_space" } }))
     expect(SessionRetry.retryable(error, retryProvider)).toBeUndefined()
@@ -436,4 +453,74 @@ describe("session.message-v2.fromError", () => {
       message: "An error occurred while processing your request.",
     })
   })
+
+  test("maps wrapped OpenAI stream errors to retryable APIError", () => {
+    const result = MessageV2.fromError(
+      new Error(
+        JSON.stringify({
+          error: {
+            code: "server_is_overloaded",
+            type: "server_error",
+            message: "The model is currently overloaded.",
+          },
+        }),
+      ),
+      { providerID: ProviderV2.ID.make("openai") },
+    )
+
+    expect(SessionV1.APIError.isInstance(result)).toBe(true)
+    if (!SessionV1.APIError.isInstance(result)) throw new Error("expected APIError")
+
+    expect(result.data.isRetryable).toBe(true)
+    expect(SessionRetry.retryable(result, retryProvider)).toBeDefined()
+  })
+
+  test("maps wrapped OpenAI stream error types with empty codes to retryable APIError", () => {
+    const result = MessageV2.fromError(
+      new Error(
+        JSON.stringify({
+          error: {
+            code: "",
+            type: "server_error",
+            message: "The upstream service failed.",
+          },
+        }),
+      ),
+      { providerID: ProviderV2.ID.make("openai") },
+    )
+
+    expect(SessionV1.APIError.isInstance(result)).toBe(true)
+    if (!SessionV1.APIError.isInstance(result)) throw new Error("expected APIError")
+
+    expect(result.data.isRetryable).toBe(true)
+    expect(SessionRetry.retryable(result, retryProvider)).toBeDefined()
+  })
+
+  test("maps double-wrapped OpenAI stream errors to retryable APIError", () => {
+    const result = MessageV2.fromError(
+      new Error(
+        JSON.stringify({
+          error: {
+            type: "error",
+            sequence_number: 2,
+            error: {
+              code: "server_is_overloaded",
+              type: "service_unavailable_error",
+              message: "Our servers are currently overloaded. Please try again later.",
+            },
+          },
+        }),
+      ),
+      { providerID: ProviderV2.ID.make("openai") },
+    )
+
+    expect(SessionV1.APIError.isInstance(result)).toBe(true)
+    if (!SessionV1.APIError.isInstance(result)) throw new Error("expected APIError")
+
+    expect(result.data.message).toBe("Our servers are currently overloaded. Please try again later.")
+    expect(result.data.isRetryable).toBe(true)
+    expect(SessionRetry.retryable(result, retryProvider)).toEqual({
+      message: "Our servers are currently overloaded. Please try again later.",
+    })
+  })
 })
diff --git a/packages/opencode/test/tool/task.test.ts b/packages/opencode/test/tool/task.test.ts
index 6a9604a01efa..18cd5c938394 100644
--- a/packages/opencode/test/tool/task.test.ts
+++ b/packages/opencode/test/tool/task.test.ts
@@ -1,7 +1,7 @@
 import { afterEach, describe, expect } from "bun:test"
 import { SessionV1 } from "@opencode-ai/core/v1/session"
 import { Database } from "@opencode-ai/core/database/database"
-import { Deferred, Effect, Exit, Fiber, Layer } from "effect"
+import { Cause, Deferred, Effect, Exit, Fiber, Layer } from "effect"
 import { Agent } from "../../src/agent/agent"
 import { BackgroundJob } from "@/background/job"
 import { EventV2Bridge } from "@/event-v2-bridge"
@@ -380,6 +380,57 @@ describe("tool.task", () => {
     }),
   )
 
+  it.instance("execute fails when child prompt returns an error without text", () =>
+    Effect.gen(function* () {
+      const { chat, assistant } = yield* seed()
+      const tool = yield* TaskTool
+      const def = yield* tool.init()
+
+      const result = yield* def
+        .execute(
+          {
+            description: "inspect bug",
+            prompt: "look into the cache key path",
+            subagent_type: "general",
+          },
+          {
+            sessionID: chat.id,
+            messageID: assistant.id,
+            agent: "build",
+            abort: new AbortController().signal,
+            extra: {
+              promptOps: {
+                ...stubOps(),
+                prompt: (input) =>
+                  Effect.sync(() => {
+                    const text = reply(input, "ok")
+                    return {
+                      ...text,
+                      info: {
+                        ...text.info,
+                        error: new SessionV1.APIError({
+                          message: "child failed",
+                          isRetryable: false,
+                        }).toObject(),
+                      },
+                      parts: [],
+                    }
+                  }),
+              } satisfies TaskPromptOps,
+            },
+            messages: [],
+            metadata: () => Effect.void,
+            ask: () => Effect.void,
+          },
+        )
+        .pipe(Effect.exit)
+
+      expect(Exit.isFailure(result)).toBe(true)
+      if (!Exit.isFailure(result)) return
+      expect(Cause.pretty(result.cause)).toContain("Child assistant failed")
+    }),
+  )
+
   it.instance(
     "execute shapes child permissions for task, todowrite, and primary tools",
     () =>