fix(llm): prevent Bedrock throttling errors from being misclassified as context overflow

packages/app-bundle/manifest.json

@@ -43,7 +43,9 @@
     "packages/core/test/location-mutation.test.ts": "2c4c8faa220cc8fabfe12d57bd74ec8dd8aac8485d0ecafd2761cdee1806d6b0",
     "packages/core/test/tool-question.test.ts": "4df41393f15104a1d3a9c3d06e30cef515108490b41fcf4d9a7d9399fd4ed945",
     "packages/llm/src/cache-policy.ts": "04c1fb5d3d061cfa95d04e54df0a1c2cdaafac921224242446eaa2968bff7857",
+    "packages/llm/src/provider-error.ts": "65b9124905f0755c1e368df5150aa211bea3cc29afb66845b5ceaccba10042dd",
     "packages/llm/test/cache-policy.test.ts": "e921095c4301440c786105063439e49b2fedbaf6d46f30457e0a96388d268f4b",
+    "packages/llm/test/provider-error.test.ts": "ac3bf099319acdc27ff47ea79fd74357d55e12ab95131a07a72a5a2a9bc0c08f",
     "packages/opencode/script/build.ts": "876e7d569aa71ce060c1d4b24c96e9bf13245cbf4ca743828d645560a95aa641",
     "packages/opencode/script/httpapi-exercise.ts": "772bd2cf5748ca90c5d5790c2d044237518cc8ede0411fc4ba28b9d7440ba28c",
     "packages/opencode/test/amicode-release-workflow.test.ts": "f6dfcb3ed9c572ca8e8b21ff82e038872b635b25c68f9580f6d6f51ee2813189",
@@ -274,7 +276,7 @@
     "packages/opencode/src/project/instance-context.ts": "65a35792e50e96a50ea14cd9f928045c0e2fc15eb49506f5bbceee38c4af1457",
     "packages/opencode/src/project/instance-store.ts": "c22d3e2499e63dd8d9395d58189df75444d306774df833e746f22e327edcf2ab",
     "packages/opencode/src/provider/auth.ts": "b0ab1322c1039e27176bf9e3500603eae27506065ac141a69710220f3361f6b8",
-    "packages/opencode/src/provider/error.ts": "7f3c8210c775e2ce98140bf015dc97e61192912af062c75e18ecb09ee2736a0f",
+    "packages/opencode/src/provider/error.ts": "d72310be26cfae5c2943f9939a6461e848532a5646ff405d628e9a50fc1ebe91",
     "packages/opencode/src/provider/provider.ts": "671f6d8de83f919ddaa69b8240cacd82e2b8202ab505d4334175f2858bbd8c32",
     "packages/opencode/src/provider/transform.ts": "d35a95a6092b5dd9cc8e37adb4d52f9ad9880d021cb3119c1d4abbc4b4fb0772",
     "packages/opencode/src/server/server.ts": "56b3c2047917aa80d1dfb29bfa9839bcc2ad7f5a6f4db12103cd962a98fba866",
@@ -344,7 +346,7 @@
     "packages/opencode/test/session/business-record.test.ts": "7857094479408bcfd6ea8d9f5b425c6384bcf8490f930674bc1c203e82782d75",
     "packages/opencode/test/session/instruction.test.ts": "8b4d954eb10afa852a671e6c0a94d19604deacc9a62f9c3651c7456cc32736b7",
     "packages/opencode/test/session/llm-native-recorded.test.ts": "7139b58569af0172862af521e8f8d7a78132f40f06dcceb4baf8ca3a2f07d3bd",
-    "packages/opencode/test/session/message-v2.test.ts": "9edc16ae57b7f2d34aaed7378d82fcf9d07a82075a05730bfb3caae39d70ac7f",
+    "packages/opencode/test/session/message-v2.test.ts": "d1b930b4b57243387920b76fb68680e507e3a288d50d06c357d83eb7931da735",
     "packages/opencode/test/session/mutation.test.ts": "ccb82b6858ac097d2af09a94497d05f6a0ff843b928a69e763fdb510905a10cd",
     "packages/opencode/test/session/prompt.test.ts": "e3d0748edb833640ef284857ae725e4b974843480d4128f56239e3fd1e6d481f",
     "packages/opencode/test/session/prose-guard.test.ts": "7239a758cbf27df4c3879fa9ee0374964559c1461773c2f5a85196f64632e6c4",
@@ -849,7 +851,7 @@
     "packages/ui/src/v2/components/text-shimmer-v2.tsx"
   ],
   "exceptions": [],
-  "extracted_at": "2026-09-21T02:29:31.939Z",
+  "extracted_at": "2026-09-21T17:30:46.454Z",
   "per_package": {
     "packages/app": {
       "A": 171,
@@ -863,7 +865,7 @@
     },
     "packages/llm": {
       "A": 0,
-      "M": 4,
+      "M": 6,
       "D": 0
     },
     "packages/opencode": {
@@ -893,7 +895,7 @@
     }
   },
   "counts": {
-    "overlay_total": 827,
+    "overlay_total": 829,
     "deletions": 7,
     "server_coupled": 0
   },
@@ -933,7 +935,9 @@
     "packages/core/test/location-mutation.test.ts": "M",
     "packages/core/test/tool-question.test.ts": "M",
     "packages/llm/src/cache-policy.ts": "M",
+    "packages/llm/src/provider-error.ts": "M",
     "packages/llm/test/cache-policy.test.ts": "M",
+    "packages/llm/test/provider-error.test.ts": "M",
     "packages/opencode/script/build.ts": "M",
     "packages/opencode/script/httpapi-exercise.ts": "M",
     "packages/opencode/test/amicode-release-workflow.test.ts": "A",

packages/app-bundle/overlay/packages/llm/src/provider-error.ts

@@ -0,0 +1,55 @@
+import { Schema } from "effect"
+import { LLMError, ProviderErrorEvent } from "./schema"
+
+const patterns = [
+  /prompt is too long/i,
+  /request_too_large/i,
+  /input is too long for requested model/i,
+  /exceeds the context window/i,
+  /exceeds (?:the )?(?:model'?s )?maximum context length(?: of [\d,]+ tokens?|\s*\([\d,]+\))/i,
+  /input token count.*exceeds the maximum/i,
+  /tokens in request more than max tokens allowed/i,
+  /maximum prompt length is \d+/i,
+  /reduce the length of the messages/i,
+  /maximum context length is \d+ tokens/i,
+  /exceeds (?:the )?maximum allowed input length of [\d,]+ tokens?/i,
+  /input \(\d+ tokens\) is longer than the model'?s context length \(\d+ tokens\)/i,
+  // Tightened: require a context/input/prompt noun so throughput/quota limits
+  // like "exceeds the limit of 100000 tokens per minute" don't false-positive.
+  /(?:input|context|prompt|message).*exceeds the limit of \d+/i,
+  /exceeds the available context size/i,
+  /greater than the context length/i,
+  /context window exceeds limit/i,
+  /exceeded model token limit/i,
+  /context[_ ]length[_ ]exceeded/i,
+  /request entity too large/i,
+  /context length is only \d+ tokens/i,
+  /input length.*exceeds.*context length/i,
+  /prompt too long; exceeded (?:max )?context length/i,
+  /too large for model with \d+ maximum context length/i,
+  /prompt has [\d,]+ tokens?, but the configured context size is [\d,]+ tokens?/i,
+  /model_context_window_exceeded/i,
+  /^too many tokens$/i,
+  /^token limit exceeded$/i,
+]
+
+const exclusions = [
+  /^(throttling error|service unavailable):/i,
+  /rate limit/i,
+  /too many requests/i,
+  // Bedrock ThrottlingException messages that match broad patterns above but
+  // are throughput/quota limits, not context-window overflow.
+  /throughput/i,
+  /per\s+(minute|second|hour|day)/i,
+  /invocation.*(limit|quota|exceeded)/i,
+  /quota/i,
+]
+
+export const isContextOverflow = (message: string) =>
+  !exclusions.some((pattern) => pattern.test(message)) &&
+  (patterns.some((pattern) => pattern.test(message)) || /^4(00|13)\s*(status code)?\s*\(no body\)/i.test(message))
+
+export const isContextOverflowFailure = (failure: unknown) =>
+  failure instanceof LLMError
+    ? failure.reason._tag === "InvalidRequest" && failure.reason.classification === "context-overflow"
+    : Schema.is(ProviderErrorEvent)(failure) && failure.classification === "context-overflow"

packages/app-bundle/overlay/packages/llm/test/provider-error.test.ts

@@ -0,0 +1,57 @@
+import { describe, expect, test } from "bun:test"
+import { isContextOverflow } from "../src"
+
+describe("provider error classification", () => {
+  test("classifies provider token limit messages as context overflow", () => {
+    const messages = [
+      "tokens in request more than max tokens allowed",
+      '{"error":{"type":"request_too_large","message":"Request exceeds the maximum size"}}',
+      "Requested token count exceeds the model's maximum context length of 131072 tokens.",
+      "Input length (265330) exceeds model's maximum context length (262144).",
+      "Input length 131393 exceeds the maximum allowed input length of 131040 tokens.",
+      "The input (516368 tokens) is longer than the model's context length (262144 tokens).",
+      "Prompt has 5,958,968 tokens, but the configured context size is 256,000 tokens",
+      "Too many tokens",
+      "Token limit exceeded",
+    ]
+
+    expect(messages.every(isContextOverflow)).toBe(true)
+  })
+
+  test("does not classify rate limits as context overflow", () => {
+    const messages = [
+      "Throttling error: Too many tokens, please wait before trying again.",
+      "Rate limit exceeded, please retry after 30 seconds.",
+      "Too many requests. Please slow down.",
+      "Monthly token limit exceeded",
+    ]
+
+    expect(messages.some(isContextOverflow)).toBe(false)
+  })
+
+  test("does not classify Bedrock throttling/quota messages as context overflow", () => {
+    const messages = [
+      // ThrottlingException messages that match broad patterns (too many tokens,
+      // token limit exceeded, exceeds the limit) but are throughput limits.
+      "Token throughput limit exceeded",
+      "Too many tokens per minute for model anthropic.claude-opus-4-8-20250808-v1:0",
+      "You have exceeded the model invocation throughput limit of 100000 tokens per minute",
+      "Model invocation throughput limit exceeded",
+      "Insufficient quota for this request",
+      "Request quota exceeded, please try again later",
+      "Token throughput quota exceeded: too many tokens per second",
+    ]
+
+    expect(messages.some(isContextOverflow)).toBe(false)
+  })
+
+  test("still classifies real Bedrock context overflow as context overflow", () => {
+    const messages = [
+      "Input is too long for requested model. Max input token count: 200000, actual token count: 250000",
+      "Input is too long for requested model",
+      "Input length 265330 exceeds the context window of 200000 tokens",
+    ]
+
+    expect(messages.every(isContextOverflow)).toBe(true)
+  })
+})

packages/app-bundle/overlay/packages/opencode/src/provider/error.ts

@@ -176,7 +176,15 @@ export type ParsedAPICallError =
 export function parseAPICallError(input: { providerID: ProviderV2.ID; error: APICallError }): ParsedAPICallError {
   const m = message(input.providerID, input.error)
   const body = json(input.error.responseBody)
-  if (isContextOverflow(m) || input.error.statusCode === 413 || body?.error?.code === "context_length_exceeded") {
+  // Status-code guard: 429 (rate limit) and 5xx (server error) are never
+  // context overflow, even when the error message text happens to match the
+  // heuristic patterns (e.g. Bedrock ThrottlingException "token throughput
+  // limit exceeded" matches /token limit exceeded/i but is a rate limit).
+  // The `undefined` status case (stream errors without HTTP metadata) still
+  // runs the heuristic — only known non-overflow statuses are excluded.
+  const status = input.error.statusCode
+  const overflowEligible = status === undefined || (status >= 400 && status < 500 && status !== 429)
+  if (overflowEligible && (isContextOverflow(m) || status === 413 || body?.error?.code === "context_length_exceeded")) {
     return {
       type: "context_overflow",
       message: m,

packages/app-bundle/overlay/packages/opencode/test/session/message-v2.test.ts

@@ -1538,6 +1538,56 @@ describe("session.message-v2.fromError", () => {
     expect(SessionV1.APIError.isInstance(result)).toBe(true)
   })
 
+  test("does not classify Bedrock 429 throttling with overflow-matching body as context overflow", () => {
+    // Bedrock's @ai-sdk/amazon-bedrock adapter coerces undefined to "undefined",
+    // stripping the SDK's "429 Too Many Requests" prefix that would otherwise
+    // trigger the "too many requests" exclusion in isContextOverflow. The status
+    // code guard in parseAPICallError must prevent the raw Bedrock message
+    // (which matches /token limit exceeded/i) from being misclassified.
+    const bedrockMessages = [
+      "Token throughput limit exceeded",
+      "Too many tokens per minute for model anthropic.claude-opus-4-8-20250808-v1:0",
+      "You have exceeded the model invocation throughput limit of 100000 tokens per minute",
+    ]
+
+    for (const bodyMessage of bedrockMessages) {
+      const result = MessageV2.fromError(
+        new APICallError({
+          message: "undefined",
+          url: "https://bedrock-runtime.us-east-1.amazonaws.com/model/anthropic.claude-opus-4-8-20250808-v1%3A0/converse-stream",
+          requestBodyValues: {},
+          statusCode: 429,
+          responseHeaders: { "content-type": "application/json" },
+          responseBody: JSON.stringify({ message: bodyMessage }),
+          isRetryable: true,
+        }),
+        { providerID: ProviderV2.ID.make("amazon-bedrock") },
+      )
+      expect(SessionV1.ContextOverflowError.isInstance(result)).toBe(false)
+      expect(SessionV1.APIError.isInstance(result)).toBe(true)
+    }
+  })
+
+  test("does not classify Bedrock 400 throttling with overflow-matching body as context overflow", () => {
+    // Bedrock ValidationException (HTTP 400) can carry throughput messages that
+    // match context-overflow patterns. The broadened exclusions in
+    // isContextOverflow must prevent these from being misclassified.
+    const result = MessageV2.fromError(
+      new APICallError({
+        message: "undefined",
+        url: "https://bedrock-runtime.us-east-1.amazonaws.com/model/anthropic.claude-opus-4-8-20250808-v1%3A0/converse-stream",
+        requestBodyValues: {},
+        statusCode: 400,
+        responseHeaders: { "content-type": "application/json" },
+        responseBody: JSON.stringify({ message: "Token throughput limit exceeded" }),
+        isRetryable: false,
+      }),
+      { providerID: ProviderV2.ID.make("amazon-bedrock") },
+    )
+    expect(SessionV1.ContextOverflowError.isInstance(result)).toBe(false)
+    expect(SessionV1.APIError.isInstance(result)).toBe(true)
+  })
+
   test("serializes unknown inputs", () => {
     const result = MessageV2.fromError(123, { providerID })