@bitkyc08/opencodex 2.87.0 → 2.88.0

package/src/server/index.ts

@@ -267,17 +267,17 @@ function startServerWithSpendLedgerOwner(port: number | undefined, deps: StartSe
     startupOwnershipHomes = homes;
     startupOwnershipStatePaths = statePaths;
   } catch { /* inspection below stays unknown */ }
-  const startupCacheOwnership = inspectStartupOwnership(
+  const startupCacheOwnership = shouldSyncCodexOnStart(config) ? inspectStartupOwnership(
     deps,
     startupOwnershipHomes,
     startupOwnershipStatePaths,
     startupWindowsTaskListingCache,
-  );
+  ) : null;
   // Startup cache invalidation is best-effort and applies only when startup sync is enabled.
   // Check OFF before taking K: on Windows K resolves SID + LocalAppData through PowerShell,
   // and run 35093667426 exceeded healthy controls by 33.8 s against that 30 s child budget.
   // The permit callback still re-reads intent under K to close a concurrent disable race.
-  if (shouldSyncCodexOnStart(config) && startupCacheOwnership.ownership === "owned" && startupOwnershipHomes !== null) {
+  if (startupCacheOwnership?.ownership === "owned" && startupOwnershipHomes !== null) {
     try {
       const startupCodexHome = startupOwnershipHomes.codexHome;
       // #1046: record whether this actually rewrote the cache. `handleStart` ORs this
@@ -549,12 +549,12 @@ function startServerWithSpendLedgerOwner(port: number | undefined, deps: StartSe
   // clients; no Codex request can use this lifecycle in that state.
   // Re-probe here instead of trusting the earlier cache decision: startup work
   // between the two sites must not widen the service-install race.
-  const nativeOwnership = inspectStartupOwnership(
+  const nativeOwnership = shouldSyncCodexOnStart(config) ? inspectStartupOwnership(
     deps,
     startupOwnershipHomes,
     startupOwnershipStatePaths,
-  );
-  const preparedNativeMainLifecycle = nativeOwnership.ownership !== "foreign"
+  ) : null;
+  const preparedNativeMainLifecycle = nativeOwnership !== null && nativeOwnership.ownership !== "foreign"
     && startupOwnershipHomes !== null
     ? prepareNativeMainStartupLifecycle(
       deps.nativeMainStartup,
@@ -600,7 +600,7 @@ function startServerWithSpendLedgerOwner(port: number | undefined, deps: StartSe
       return retryPreparedNativeMainLifecycle.start();
     },
   };
-  const nativeMainLifecycle: NativeMainStartupLifecycle = shouldSyncCodexOnStart(config)
+  const nativeMainLifecycle: NativeMainStartupLifecycle = nativeOwnership !== null
     ? nativeOwnership.ownership === "owned"
       ? preparedNativeMainLifecycle
         ? preparedNativeMainLifecycle.start()

package/src/server/lifecycle.ts

@@ -38,6 +38,7 @@ export interface ActiveTurnLease extends AdmissionLease {
   attach(lease: AdmissionLease): void;
   bindAbortSignal(signal: AbortSignal): () => void;
   bindAbortController(ac: AbortController): void;
+  unbindAbortController(ac: AbortController): void;
   beginCodexAccountSelection(): CodexAccountSelectionAdmission;
   isTransferred(): boolean;
 }
@@ -231,6 +232,17 @@ export function tryAdmitTurn(sessionLaneId?: string, clientSignal?: AbortSignal)
       activeTurns.set(ac, lease);
       lease.bindAbortSignal(ac.signal);
     },
+    unbindAbortController(ac) {
+      if (!controllers.delete(ac)) return;
+      if (activeTurns.get(ac) === lease) activeTurns.delete(ac);
+      const listener = abortListeners.get(ac.signal);
+      if (listener) {
+        ac.signal.removeEventListener("abort", listener);
+        abortListeners.delete(ac.signal);
+      }
+      // A completed child relay cannot retire its still-running parent turn.
+      if (controllers.size === 0) lease.release();
+    },
     beginCodexAccountSelection() {
       const mainProfileDraining = nativeMainDrainOwners.size > 0;
       let selectionActive = active && !mainProfileDraining;
@@ -327,7 +339,7 @@ export function unregisterTurn(ac: AbortController): void {
     turnReleaseMisses += 1;
     return;
   }
-  lease.release();
+  lease.unbindAbortController(ac);
 }
 export function isDraining(): boolean { return shutdownDraining || temporaryDrainOwners.size > 0; }
 export function getActiveTurnCount(): number { return turnGate.metrics().active; }

package/src/server/messages-native.ts

@@ -44,6 +44,7 @@ import { classifyError } from "../lib/errors";
 import { relaySseWithPayloadRewrite } from "./sse-payload-rewrite";
 import { redactSecretString } from "../lib/redact";
 import { resolveClientRetryAfter } from "../lib/retry-after";
+import { enrichOpenCodeZenUpstreamMessage } from "../providers/opencode-zen-rate-limit";
 import { isTranslatorBudgetExceededError, type TranslatorBudget } from "../lib/translator-budget";
 import {
   applyReplayRefusalClientHeaders,
@@ -845,7 +846,7 @@ export async function handleNativeMessages(options: HandleNativeMessagesOptions)
       upstream.abort();
       return fail(499, "Client cancelled request", "api_error");
     }
-    return retainUpstreamMessagesRequestId(nativeMessagesErrorResponse(response, bodyText, finishLog), response.headers);
+    return retainUpstreamMessagesRequestId(nativeMessagesErrorResponse(response, bodyText, finishLog, route.providerName, activeProvider), response.headers);
   }
 
   const contentType = response.headers.get("content-type")?.toLowerCase() ?? "";
@@ -999,7 +1000,7 @@ function isNativeOverflowError(error: Rec, message: string): boolean {
  * Messages path, so a client sees the same contract on either lane: transient 5xx become 529
  * with a retry hint, and a proxy-side replay refusal keeps its no-retry marking.
  */
-function nativeMessagesErrorResponse(response: Response, bodyText: string, finishLog: FinishLog): Response {
+function nativeMessagesErrorResponse(response: Response, bodyText: string, finishLog: FinishLog, providerName: string, provider: OcxProviderConfig): Response {
   let upstreamType: string | undefined;
   let upstreamCode: string | undefined;
   let upstreamMessage: string | undefined;
@@ -1013,8 +1014,20 @@ function nativeMessagesErrorResponse(response: Response, bodyText: string, finis
     }
   } catch { /* keep generic classification */ }
   const detail = formatAnthropicErrorBody(response.status, response.headers, bodyText);
-  const message = upstreamMessage
+  const rawMessage = upstreamMessage
     ?? (detail ? `Provider error ${response.status}: ${detail}` : `Provider error ${response.status}`);
+  const message = enrichOpenCodeZenUpstreamMessage(rawMessage, {
+    status: response.status,
+    providerName,
+    baseUrl: provider.baseUrl,
+    adapter: provider.adapter,
+    authMode: provider.authMode,
+    hasApiKey: Boolean(provider.apiKey?.trim()),
+    upstreamRetryAfter: response.headers.get("retry-after"),
+    upstreamErrorType: upstreamType,
+    upstreamErrorCode: upstreamCode,
+    supportsHttpSameKeyRetry: true,
+  });
   const classified = classifyError(
     response.status,
     upstreamType ?? (response.status === 401 ? "authentication_error"

package/src/server/responses/adapter-continuation.ts

@@ -1,3 +1,4 @@
+import { bindAnthropicRejectionResponse } from "../../adapters/anthropic-rejection-memo";
 import { rotateAnthropicAccountOnResponseForInstance } from "../../oauth/anthropic-account-refusal";
 import type { ResponsesRequestContext } from "./core-options";
 import type { PreparedResponsesRequest } from "./request-prepare";
@@ -29,6 +30,7 @@ import {
   applyUpstreamRecoveryInit,
   isNonReplayableResponse,
   prepareSameTarget429Wait,
+  SendBudgetExhaustedError,
 } from "../../lib/upstream-retry";
 import { unwrapUpstreamRetryEvidenceError } from "../../codex/auth-context";
 import { AdmissionModelDeniedError, MODEL_NOT_ALLOWED_FOR_KEY } from "../admission-model-scope";
@@ -121,6 +123,7 @@ export function createAdapterContinuations(
     | "rateLimitRetries"
     | "stallTimeoutMs"
     | "localUpstream"
+    | "thinkingSignatureRecovery"
    >,
 ) {
   const { options, logCtx, config } = requestContext;
@@ -159,6 +162,10 @@ export function createAdapterContinuations(
         errorType: MODEL_NOT_ALLOWED_FOR_KEY, code: MODEL_NOT_ALLOWED_FOR_KEY, retryable: false,
       };
     }
+    if (refusal instanceof SendBudgetExhaustedError) return {
+      type: "error", status: 429, errorType: "rate_limit_error", code: refusal.code,
+      message: refusal.message, retryable: false,
+    };
     return {
       type: "error",
       message: `Provider continuation failed: ${redactSecretString(error instanceof Error ? error.message : String(error))}`,
@@ -309,6 +316,14 @@ export function createAdapterContinuations(
         const recoveryKind = nextContinuationRecoveryKind;
         nextContinuationRecoveryKind = undefined;
         response = await fetchContinuation(recoveryKind);
+        bindAnthropicRejectionResponse(transportState.sameTargetRequest, response);
+        if (initialRecoveryKind === "empty-completion" && await adapterExchange.thinkingSignatureRecovery.prepareHttp(
+          response, transportState.activeAdapter.name, nextParsed,
+        )) {
+          try { void response.body?.cancel().catch(() => {}); } catch { /* closed */ }
+          invalidateSameTargetRequest();
+          response = await fetchContinuation("opaque-blob-rejection");
+        }
         if (kiroRefusalPendingReplay) {
           try { void kiroRefusalPendingReplay.body?.cancel().catch(() => {}); } catch { /* already closed */ }
           kiroRefusalPendingReplay = undefined;
@@ -704,24 +719,25 @@ export function createAdapterContinuations(
       // listener that can cancel its reader even while the parser holds the body lock.
       // The shared signal is not aborted on timeout: this generator must still emit
       // the error that the already-live Responses bridge turns into a terminal.
-      if (nextParsed.stream) {
-        yield* readResponseStreamWithInactivity(
-          response,
-          upstream.signal,
-          bodyInactivityMs,
-          guarded => transportState.activeAdapter.parseStream(guarded, translatorBudget, logCtx.activeTierMetadata),
-        );
-      } else if (transportState.activeAdapter.parseResponse) {
-        yield* await readResponseBodyWithInactivity(
-          response,
-          upstream.signal,
-          bodyInactivityMs,
-          guarded => transportState.activeAdapter.parseResponse!(guarded, translatorBudget, logCtx.activeTierMetadata),
-        );
-      } else {
-        try { void response.body?.cancel().catch(() => {}); } catch { /* already closed */ }
-        yield { type: "error", message: "Provider continuation does not support response parsing" };
-      }
+      const parse = async function* (current: Response): AsyncGenerator<AdapterEvent> {
+        if (!current.ok) {
+          yield await adapterExchange.thinkingSignatureRecovery.projectHttpFailure(current, upstream.signal);
+        } else if (nextParsed.stream) {
+          yield* readResponseStreamWithInactivity(current, upstream.signal, bodyInactivityMs,
+            guarded => transportState.activeAdapter.parseStream(guarded, translatorBudget, logCtx.activeTierMetadata));
+        } else if (transportState.activeAdapter.parseResponse) {
+          yield* await readResponseBodyWithInactivity(current, upstream.signal, bodyInactivityMs,
+            guarded => transportState.activeAdapter.parseResponse!(guarded, translatorBudget, logCtx.activeTierMetadata));
+        } else {
+          try { void current.body?.cancel().catch(() => {}); } catch { /* closed */ }
+          yield { type: "error", message: "Provider continuation does not support response parsing" };
+        }
+      };
+      yield* adapterExchange.thinkingSignatureRecovery.wrap(parse(response), async function* () {
+        invalidateSameTargetRequest();
+        yield* parse(await fetchContinuation("opaque-blob-rejection"));
+      }, () => transportState.activeAdapter.name,
+      initialRecoveryKind === "empty-completion" && !isNonReplayableResponse(response), response, nextParsed);
     } catch (error) {
       // Classify the thrown error before reading signal state: cancelling the stalled
       // source can synchronously abort the shared signal, which would otherwise report

package/src/server/responses/adapter-delivery.ts

@@ -23,6 +23,9 @@ import { resolveStallTimeoutMs } from "../../stall-timeout";
 import { clientEncoderForDelivery, deliverClientEncodedResponse } from "../inference/client-encoder-delivery";
 import { noteKiroServedSuccess } from "../../providers/kiro-usage";
 import { persistKiroAccountState } from "../../providers/kiro-account-state-disk";
+import { readDisplaySafeErrorText, normalizeUpstreamErrorText } from "./core-errors";
+import { sanitizeNonReplayableUpstreamError } from "./non-replayable-error";
+import { isNonReplayableResponse } from "../../lib/upstream-retry";
 import { isCanonicalOpenAiForwardProvider } from "../../providers/openai-tiers";
 import { guardEmptyToolCatalogEvents, guardEmptyToolCatalogStream } from "../../lib/empty-tool-catalog";
 
@@ -49,7 +52,7 @@ export async function deliverAdapterResponse(
     | "notifyResponseComplete"
   >,
   completionPolicy: Pick<ResponsesCompletionPolicy, "emptyCompletionGuardEnabled">,
-  adapterExchange: Pick<AdapterExchange, "upstreamResponse" | "upstream" | "cleanupUpstreamAbort" | "localUpstream">,
+  adapterExchange: Pick<AdapterExchange, "upstreamResponse" | "upstream" | "cleanupUpstreamAbort" | "localUpstream" | "thinkingSignatureRecovery" | "refetchThinkingSignature">,
   continuationState: Pick<AdapterContinuations, "terminalGuardEnabled" | "fetchTerminalGuardContinuation" | "fetchGuardedEmptyCompletionRetry">,
 ): Promise<Response> {
   const { logCtx, options, config } = requestContext;
@@ -61,7 +64,7 @@ export async function deliverAdapterResponse(
     rememberKiroDeliveredFinalAnswer,
     responseStateOptions,
   } = requestState;
-  const { upstreamResponse, upstream, cleanupUpstreamAbort, localUpstream } = adapterExchange;
+  const { upstreamResponse, upstream, cleanupUpstreamAbort, localUpstream, thinkingSignatureRecovery, refetchThinkingSignature } = adapterExchange;
   const {
     terminalGuardEnabled,
     fetchTerminalGuardContinuation,
@@ -77,6 +80,43 @@ export async function deliverAdapterResponse(
   const { routedCompaction } = sidecarState;
   const bodyInactivityMs = resolveStallTimeoutMs(config.stallTimeoutSec, { localUpstream });
   const upstreamRequestsStream = parsed.stream || isCanonicalOpenAiForwardProvider(route.provider);
+  const recoverThinkingSignature = (events: AsyncIterable<AdapterEvent>): AsyncIterable<AdapterEvent> =>
+    thinkingSignatureRecovery.wrap(events, async function* () {
+      const result = await refetchThinkingSignature();
+      let response = "failed" in result ? result.failed : result;
+      // A replacement can be an HTTP/auth/budget refusal; never feed its JSON body to an SSE parser.
+      if (!response.ok) {
+        // A local refusal closes the upstream signal before returning its generated JSON.
+        // Reading that bounded local body still honors caller cancellation.
+        const errorSignal = "failed" in result ? options.abortSignal ?? new AbortController().signal : upstream.signal;
+        const nonReplayable = isNonReplayableResponse(response);
+        if (nonReplayable) response = await sanitizeNonReplayableUpstreamError(response, errorSignal);
+        const body = await readDisplaySafeErrorText(response, errorSignal, "Upstream recovery failed");
+        const error = normalizeUpstreamErrorText(body, "Upstream recovery failed");
+        yield { type: "error", status: response.status, message: error.message ?? error.safeText,
+          errorType: error.type ?? "upstream_error", ...(error.code ? { code: error.code } : {}),
+          ...(nonReplayable ? { retryable: false } : {}) };
+        return;
+      }
+      try {
+        const retryEvents = upstreamRequestsStream
+          ? readResponseStreamWithInactivity(response, upstream.signal, bodyInactivityMs,
+            guarded => transportState.activeAdapter.parseStream(guarded, translatorBudget, logCtx.activeTierMetadata))
+          : (async function* () {
+              yield* await readResponseBodyWithInactivity(response, upstream.signal, bodyInactivityMs,
+                guarded => transportState.activeAdapter.parseResponse!(guarded, translatorBudget, logCtx.activeTierMetadata));
+            })();
+        for await (const event of retryEvents) {
+          options.onCompactionRecoveryAdapterEvent?.(event);
+          yield event;
+        }
+      } catch (error) {
+        if (!(error instanceof ResponseBodyInactivityError)) throw error;
+        yield { type: "error", message: "Upstream recovery response body stalled before completing",
+          status: 504, errorType: "upstream_error" };
+      }
+    }, () => transportState.activeAdapter.name, !isNonReplayableResponse(upstreamResponse), upstreamResponse);
+
   const clientToolCatalog = requestState.clientToolCatalogRestriction && route.provider.authMode !== "forward"
     ? requestState.clientToolCatalogRestriction : undefined;
   const deliveryToolBridgeMaps = clientToolCatalog
@@ -110,15 +150,16 @@ export async function deliverAdapterResponse(
         throw error;
       }
     })();
+    const recoveredEvents = recoverThinkingSignature(initialEventStream);
     const eventStream = terminalGuardEnabled
       ? guardTerminalEventStream({
           parsed,
-          firstEvents: initialEventStream,
+          firstEvents: recoveredEvents,
           adapterName: transportState.activeAdapter.name,
           maxAutoContinuations: 1,
           continuation: next => fetchTerminalGuardContinuation(next, undefined, !parsed.stream),
         })
-      : initialEventStream;
+      : recoveredEvents;
     // Optional advisor guard (registered per request by the sidecar planner; absent unless the
     // user enabled the advisor): holds synthetic `advisor` tool calls, consults the configured
     // expert model, and re-dispatches the worker. Sits inside the empty-completion guard so an
@@ -147,6 +188,7 @@ export async function deliverAdapterResponse(
     // One completion owner for both deliveries: the bridge calls it from its terminal, the
     // direct client encoder from the fold of the same events.
     const onCompletedResponse = (response: Record<string, unknown>, providerState?: OcxProviderContinuationState) => {
+      if (!options.abortSignal?.aborted) thinkingSignatureRecovery.rememberSuccess(response);
       const served = transportState.sentOAuthSnapshot;
       if (transportState.activeAdapter.name === "kiro" && response.status === "completed" && served
         && noteKiroServedSuccess(served.accountId, served.generation)) persistKiroAccountState();
@@ -250,18 +292,22 @@ export async function deliverAdapterResponse(
         ));
       }
       for (const event of initialEvents) options.onCompactionRecoveryAdapterEvent?.(event);
+      const recoveredEvents: AdapterEvent[] = [];
+      for await (const event of recoverThinkingSignature((async function* () { yield* initialEvents; })())) {
+        recoveredEvents.push(event);
+      }
       let guardedEvents: AdapterEvent[];
       if (terminalGuardEnabled) {
         guardedEvents = [];
         for await (const event of guardTerminalEventStream({
           parsed,
-          firstEvents: (async function* () { yield* initialEvents; })(),
+          firstEvents: (async function* () { yield* recoveredEvents; })(),
           adapterName: transportState.activeAdapter.name,
           maxAutoContinuations: 1,
           continuation: next => fetchTerminalGuardContinuation(next, undefined, !parsed.stream),
         })) guardedEvents.push(event);
       } else {
-        guardedEvents = initialEvents;
+        guardedEvents = recoveredEvents;
       }
       // Optional advisor guard — same semantics as the streaming branch above.
       if (parsed._advisorGuard) {
@@ -325,6 +371,7 @@ export async function deliverAdapterResponse(
     // #1926 gap 2: same buffered-path durability bound as the primary branch.
     await awaitThoughtSignatureDurability();
     if (adapterResponseReachedServingTerminal(events, json)) {
+      if (!options.abortSignal?.aborted) thinkingSignatureRecovery.rememberSuccess(json);
       const served = transportState.sentOAuthSnapshot;
       if (transportState.activeAdapter.name === "kiro" && json.status === "completed" && served
         && noteKiroServedSuccess(served.accountId, served.generation)) persistKiroAccountState();

package/src/server/responses/adapter-dispatch.ts

@@ -1,3 +1,4 @@
+import { bindAnthropicRejectionResponse } from "../../adapters/anthropic-rejection-memo";
 import { deferPhysicalSendReporter } from "./deferred-send-reporter";
 import { spendLedgerStorageErrorResponse } from "./spend-storage-error";
 import { classifyAnthropic429 } from "../../oauth/anthropic-rate-limit-policy";
@@ -87,6 +88,7 @@ import {
   reasoningEffortRejectionText,
   anthropicFastRefused,
 } from "./core-opaque-recovery";
+import { createAnthropicThinkingRecovery } from "./anthropic-thinking-recovery";
 import { shouldAttemptImageTierRetry } from "../image-retry";
 import {
   isTransientConsoleGoUploadRejection,
@@ -591,7 +593,7 @@ export async function prepareAdapterExchange(
   let rateLimitRetries = 0;
   // Shared with the terminal-guard continuation below: an image-tier reduction that let the
   // main request clear a 413 must not be forgotten on the very next continuation build.
-  if (!upstreamResponse.ok) {
+  // Recovery helpers also serve initial HTTP-200 adapter errors during delivery.
     // Recovery loop: multi-key 429 failover + at most ONE opaque-state rebuild and ONE
     // anthropic 413 tightened retry
     // (devlog/260714_image_normalization_pipeline/030). One mutable activeAdapter serves
@@ -600,6 +602,9 @@ export async function prepareAdapterExchange(
     // 413→429 rotation cannot silently undo the tightening.
     let imageRetryAttempted = false;
     const opaqueBlobRecoveryGuard: OpaqueBlobRecoveryGuard = { attempted: false };
+    const thinkingSignatureRecovery = createAnthropicThinkingRecovery(
+      parsed, upstream.signal, opaqueBlobRecoveryGuard, () => !sendBudgetExhausted(),
+    );
     // Console Go answers a transient 400 "Invalid upload request." for bodies it accepts
     // moments later; at most one byte-identical replay is allowed per request.
     const consoleGoUploadRetryGuard: { attempted: boolean } = { attempted: false };
@@ -863,6 +868,11 @@ export async function prepareAdapterExchange(
         return { failed: formatErrorResponse(502, "upstream_error", msg) };
       }
     };
+    const refetchThinkingSignature = () => {
+      invalidateSameTargetRequest();
+      return rebuildAndRefetch("opaque-blob-rejection");
+    };
+  if (!upstreamResponse.ok) {
     const rotateAntigravityAuth = async (
       failedResponse: Response,
       recovery: AttemptRecoveryKind,
@@ -920,6 +930,17 @@ export async function prepareAdapterExchange(
           return await sanitizeNonReplayableUpstreamError(upstreamResponse, upstream.signal);
         } finally { cleanupUpstreamAbort(); }
       }
+      bindAnthropicRejectionResponse(transportState.sameTargetRequest, upstreamResponse);
+      if (await thinkingSignatureRecovery.prepareHttp(
+        upstreamResponse, transportState.activeAdapter.name,
+      )) {
+        try { void upstreamResponse.body?.cancel().catch(() => {}); } catch { /* closed */ }
+        invalidateSameTargetRequest();
+        const result = await rebuildAndRefetch("opaque-blob-rejection");
+        if ("failed" in result) return result.failed;
+        upstreamResponse = result;
+        continue recovery;
+      }
      if (
        upstreamResponse.status === 401
        && isOAuth401ReplayProvider
@@ -1877,6 +1898,8 @@ export async function prepareAdapterExchange(
             authMode: route.provider.authMode,
             hasApiKey: Boolean(route.provider.apiKey?.trim()),
             upstreamRetryAfter,
+            upstreamErrorType: normalized.type,
+            upstreamErrorCode: normalized.code,
             // This recovery path is the HTTP Responses wire; custom runTurn transports
             // never reach enrichOpenCodeZenUpstreamMessage here.
             supportsHttpSameKeyRetry: true,
@@ -1902,6 +1925,7 @@ export async function prepareAdapterExchange(
     }
   }
 
+  bindAnthropicRejectionResponse(transportState.sameTargetRequest, upstreamResponse);
   cancelBodyOnAbort(upstreamResponse.body, upstream.signal);
 
   return {
@@ -1911,6 +1935,8 @@ export async function prepareAdapterExchange(
     stallTimeoutMs,
     localUpstream,
     upstreamResponse,
+    thinkingSignatureRecovery,
+    refetchThinkingSignature,
     rateLimitPolicy,
     get rateLimitRetries(): typeof rateLimitRetries {
       return rateLimitRetries;

package/src/server/responses/anthropic-thinking-recovery.ts

@@ -0,0 +1,153 @@
+import { anthropicRejectionPolicy, bindAnthropicRejectionPolicy, rejectedAnthropicPayloads } from "../../adapters/anthropic-rejection-memo";
+import { normalizeUpstreamErrorText, readDisplaySafeErrorText } from "./core-errors";
+import { sanitizeNonReplayableUpstreamError } from "./non-replayable-error";
+import { isLikelyRealAnthropicThinkingSignature } from "../../adapters/anthropic";
+import type { AdapterEvent, OcxParsedRequest } from "../../types";
+import { upstreamErrorMessageFromPayload } from "../../lib/errors";
+import { readBoundedResponseBody } from "../../lib/bounded-body";
+import { isNonReplayableResponse } from "../../lib/upstream-retry";
+import { rememberReasoningReplayOpaqueBlobRejection } from "../../responses/reasoning-replay-cache";
+import { prepareOpaqueBlobRecovery, type OpaqueBlobRecoveryGuard } from "./core-opaque-recovery";
+
+/** Only the explicit conversation-binding refusal authorizes removing signed history. */
+export function isAnthropicThinkingSignatureBindingRejectionMessage(message: string): boolean {
+  return message.includes("Invalid `signature` in `thinking` block")
+    && message.includes("The block is bound to a different conversation");
+}
+
+function carriesThinkingState(parsed: OcxParsedRequest): boolean {
+  return parsed.context.messages.some(message => message.role === "assistant"
+    && message.content.some(part => part.type === "thinking"
+      && (isLikelyRealAnthropicThinkingSignature(part.signature)
+        || part.redacted?.some(data => typeof data === "string" && data.length > 0))));
+}
+
+/** One request owns both the HTTP refusal and initial-event recovery, never a fresh retry guard. */
+export function createAnthropicThinkingRecovery(
+  parsed: OcxParsedRequest,
+  signal: AbortSignal,
+  guard: OpaqueBlobRecoveryGuard,
+  canRetry: () => boolean,
+) {
+  let rejectedScope: OcxParsedRequest["_reasoningReplayScope"];
+  let rejectedPayloads: ReadonlySet<string> | undefined;
+  let armed = false;
+  let outputSeen = false;
+  let completed = false;
+  let failed = false;
+  let remembered = false;
+
+  const prepare = (message: string, adapterName: string, attempt = parsed, response?: Response): boolean => {
+    if (signal.aborted || !canRetry() || adapterName !== "anthropic" || guard.attempted
+      || parsed._stripReasoningEncryptedContent === true || !carriesThinkingState(attempt)
+      || !isAnthropicThinkingSignatureBindingRejectionMessage(message)) return false;
+    rejectedPayloads = rejectedAnthropicPayloads(response);
+    // A real send must prove that it serialized opaque history; otherwise rebuilding could
+    // repeat the same request. Direct event-policy callers have no HTTP body to identify.
+    if (response && !rejectedPayloads?.size) return false;
+    const previous = anthropicRejectionPolicy(parsed);
+    if (rejectedPayloads && previous?.mode === "selective") {
+      const merged = new Set(previous.fingerprints);
+      for (const payload of rejectedPayloads) merged.add(payload);
+      rejectedPayloads = merged.size <= 256 ? merged : undefined;
+    }
+    guard.attempted = true;
+    armed = true;
+    const scope = parsed._reasoningReplayScope;
+    rejectedScope = scope ? { ...scope, current: scope.current ? { ...scope.current } : undefined } : undefined;
+    if (!rejectedPayloads) {
+      prepareOpaqueBlobRecovery(parsed);
+      if (attempt !== parsed) prepareOpaqueBlobRecovery(attempt);
+    }
+    const policy = rejectedPayloads ? { mode: "selective" as const, fingerprints: rejectedPayloads } : { mode: "blanket" as const };
+    bindAnthropicRejectionPolicy(parsed, policy);
+    bindAnthropicRejectionPolicy(attempt, policy);
+    return true;
+  };
+
+  const prepareHttp = async (response: Response, adapterName: string, attempt = parsed): Promise<boolean> => {
+    if (signal.aborted || !canRetry() || response.status !== 400 || isNonReplayableResponse(response)
+      || adapterName !== "anthropic" || guard.attempted
+      || parsed._stripReasoningEncryptedContent === true || !carriesThinkingState(attempt)) return false;
+    try {
+      const body = await readBoundedResponseBody(response.clone(), { signal });
+      if (!body.displaySafe || body.truncated) return false;
+      const message = upstreamErrorMessageFromPayload(JSON.parse(body.text) as unknown);
+      return typeof message === "string" && prepare(message, adapterName, attempt, response);
+    } catch {
+      return false;
+    }
+  };
+
+  const observe = (event: AdapterEvent): void => {
+    if (!armed) return;
+    if (event.type === "error" || event.type === "incomplete") failed = true;
+    if (event.type === "done") completed = true;
+    if ((event.type === "text_delta" && event.text.length > 0) || event.type === "tool_call_end") outputSeen = true;
+  };
+
+  async function* wrap(
+    events: AsyncIterable<AdapterEvent>,
+    retry: () => AsyncIterable<AdapterEvent>,
+    adapterName: () => string,
+    replayAllowed = true,
+    response?: Response,
+    attempt = parsed,
+  ): AsyncGenerator<AdapterEvent> {
+    let source = events;
+    let replaySafe = replayAllowed;
+    while (true) {
+      const iterator = source[Symbol.asyncIterator]();
+      let recovered = false;
+      try {
+        while (true) {
+          const next = await iterator.next();
+          if (next.done) return;
+          const event = next.value;
+          if (replaySafe && event.type === "error" && prepare(event.message, adapterName(), attempt, response)) {
+            // Close the rejected body before acquiring another dispatch/pacing slot.
+            await iterator.return?.();
+            source = retry();
+            recovered = true;
+            break;
+          }
+          // Even hidden reasoning, boundaries, and tool activity fence a replay. A heartbeat
+          // carrying a side-effect latch is never evidence that a turn is safe to repeat.
+          replaySafe &&= event.type === "heartbeat" && event.replayUnsafe !== true;
+          observe(event);
+          yield event;
+        }
+      } finally {
+        if (!recovered) await iterator.return?.();
+      }
+    }
+  }
+
+  const rememberSuccess = (response: Readonly<Record<string, unknown>>): void => {
+    // HTTP 200, EOF, or a guard-generated completion alone cannot prove the stripped send worked.
+    // The delivery owner calls this only after its bridge/fold validated a successful terminal.
+    if (!armed || remembered || signal.aborted || failed || !completed || !outputSeen
+      || response.status !== "completed" || !Array.isArray(response.output) || response.output.length === 0
+      || !rejectedScope
+      || rejectedScope.clientThreadId !== parsed._reasoningReplayScope?.clientThreadId
+      || JSON.stringify(rejectedScope.current) !== JSON.stringify(parsed._reasoningReplayScope?.current)) return;
+    remembered = true;
+    rememberReasoningReplayOpaqueBlobRejection(rejectedScope, rejectedPayloads);
+  };
+  // The search owner supplies only a fully validated current iteration, never search-card output.
+  const observeIteration = (events: readonly AdapterEvent[]): void => {
+    outputSeen = false;
+    completed = false;
+    for (const event of events) observe(event);
+  };
+  const projectHttpFailure = async (response: Response, abortSignal: AbortSignal): Promise<Extract<AdapterEvent, { type: "error" }>> => {
+    const nonReplayable = isNonReplayableResponse(response);
+    if (nonReplayable) response = await sanitizeNonReplayableUpstreamError(response, abortSignal);
+    const text = await readDisplaySafeErrorText(response, abortSignal, `Provider error ${response.status}`);
+    const error = normalizeUpstreamErrorText(text, `Provider error ${response.status}`);
+    return { type: "error", status: response.status, message: error.message ?? error.safeText,
+      errorType: error.type ?? "upstream_error", ...(error.code ? { code: error.code } : {}),
+      ...(nonReplayable ? { retryable: false } : {}) };
+  };
+  return { prepareHttp, prepareEvent: prepare, observeIteration, projectHttpFailure, wrap, rememberSuccess };
+}

package/src/server/responses/combo-stream-preflight.ts

@@ -4,6 +4,7 @@ import { comboFailureDecision } from "../../combos";
 import { httpStatusFromTerminalError } from "../../lib/errors";
 import type { RequestFailureStage } from "../../lib/request-failure-model";
 import type { RequestLogContext } from "../request-log";
+import { carryResponseToolOwnership } from "./response-tool-ownership";
 import { createSseInspector } from "../relay";
 import { MAX_CLIENT_SSE_FRAME_BYTES } from "../sse-frame-buffer";
 
@@ -198,11 +199,11 @@ function replayBufferedResponse(
       reader.cancel(reason).catch(() => undefined);
     },
   });
-  return new Response(body, {
+  return carryResponseToolOwnership(response, new Response(body, {
     status: response.status,
     statusText: response.statusText,
     headers: response.headers,
-  });
+  }));
 }
 
 function failedTerminalResponse(

package/src/server/responses/compaction-recovery.ts

@@ -1,4 +1,5 @@
 import type { AdapterEvent, OcxConfig } from "../../types";
+import { carryResponseToolOwnership } from "./response-tool-ownership";
 import { routeConcreteModel, type RouteResult } from "../../router";
 import { copyPlainData } from "../../lib/plain-data";
 import { jsonUtf8Bytes } from "../../lib/json-byte-size";
@@ -78,6 +79,7 @@ async function bufferedJson(response: Response, signal: AbortSignal): Promise<{
   headers.delete("content-length");
   headers.delete("content-encoding");
   const replacement = new Response(bytes.bytes, { status: response.status, statusText: response.statusText, headers });
+  carryResponseToolOwnership(response, replacement);
   if (isNonReplayableResponse(response)) markResponseNonReplayable(replacement);
   try {
     const json: unknown = JSON.parse(new TextDecoder("utf-8", { fatal: true }).decode(bytes.bytes));

package/src/server/responses/core-codex-account.ts

@@ -975,8 +975,8 @@ export async function retryCodexPoolOnAlternateAccount(
   const maxRetrySends = retrySameConfirmedAccount ? 7 : 1;
   let retrySendCount = 0;
   // The gated-model 400 has no roster kind; it stays unattributed as before.
-  const moveRecovery: AttemptRecoveryKind | undefined = outcomeStatus === 429 || outcomeStatus === 402
-    ? "oauth-account-429"
+  const moveRecovery: AttemptRecoveryKind | undefined = outcomeStatus === 429 ? "oauth-account-429"
+    : outcomeStatus === 402 ? "oauth-account-402"
     : isTransientUpstreamStatus(outcomeStatus) ? "transient-5xx" : undefined;
   let upstreamResponse: Response;
   try {