@bitkyc08/opencodex 2.84.0 → 2.87.0

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

@@ -24,6 +24,7 @@ import { clientEncoderForDelivery, deliverClientEncodedResponse } from "../infer
 import { noteKiroServedSuccess } from "../../providers/kiro-usage";
 import { persistKiroAccountState } from "../../providers/kiro-account-state-disk";
 import { isCanonicalOpenAiForwardProvider } from "../../providers/openai-tiers";
+import { guardEmptyToolCatalogEvents, guardEmptyToolCatalogStream } from "../../lib/empty-tool-catalog";
 
 /** One responsibility of the Responses request pipeline; state owners are explicit. */
 export async function deliverAdapterResponse(
@@ -34,6 +35,7 @@ export async function deliverAdapterResponse(
     | "route"
     | "translatorBudget"
     | "toolBridgeMaps"
+    | "clientToolCatalogRestriction"
     | "rememberKiroDeliveredFinalAnswer"
     | "responseStateOptions"
   >,
@@ -75,6 +77,10 @@ export async function deliverAdapterResponse(
   const { routedCompaction } = sidecarState;
   const bodyInactivityMs = resolveStallTimeoutMs(config.stallTimeoutSec, { localUpstream });
   const upstreamRequestsStream = parsed.stream || isCanonicalOpenAiForwardProvider(route.provider);
+  const clientToolCatalog = requestState.clientToolCatalogRestriction && route.provider.authMode !== "forward"
+    ? requestState.clientToolCatalogRestriction : undefined;
+  const deliveryToolBridgeMaps = clientToolCatalog
+    ? { ...toolBridgeMaps, ...clientToolCatalog } : toolBridgeMaps;
 
 
   if (parsed.stream) {
@@ -113,17 +119,31 @@ export async function deliverAdapterResponse(
           continuation: next => fetchTerminalGuardContinuation(next, undefined, !parsed.stream),
         })
       : initialEventStream;
+    // 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
+    // advisor-only turn never reads as an empty completion, and outside the bridge so the
+    // synthetic call never reaches the client.
+    const advisorStream = parsed._advisorGuard
+      ? parsed._advisorGuard({
+          parsed,
+          translatorBudget,
+          firstEvents: eventStream,
+          continuation: fetchTerminalGuardContinuation,
+        })
+      : eventStream;
     // The empty-completion guard sits OUTSIDE the terminal guard: a completed
     // turn with no text and no tool call is retried with the IDENTICAL request
     // (fetchTerminalGuardContinuation(parsed) replays the cached byte-identical
     // request — same body, same headers, same signal).
-    const guardedEventStream = emptyCompletionGuardEnabled
+    const completionGuardedEventStream = emptyCompletionGuardEnabled
       ? guardEmptyCompletionEventStream({
-          firstEvents: eventStream,
+          firstEvents: advisorStream,
           continuation: fetchGuardedEmptyCompletionRetry,
         })
-      : eventStream;
-    const { toolNsMap, declaredToolNames, toolParameterSchemas, freeformToolNames, bareCustomToolNames, toolSearchToolNames } = toolBridgeMaps;
+      : advisorStream;
+    const guardedEventStream = guardEmptyToolCatalogStream(completionGuardedEventStream, clientToolCatalog, translatorBudget);
+    const { toolNsMap, declaredToolNames, toolParameterSchemas, freeformToolNames, bareCustomToolNames, toolSearchToolNames } = deliveryToolBridgeMaps;
     // 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) => {
@@ -145,7 +165,9 @@ export async function deliverAdapterResponse(
       }
       notifyResponseComplete(response);
     };
-    const clientEncoder = clientEncoderForDelivery(options, logCtx, !!routedCompaction, transportState.activeAdapter.name);
+    // The Responses bridge owns code-mode helper compilation; the direct encoder only renames.
+    const clientEncoder = clientToolCatalog?.bareCustomToolNames.has("exec") ? undefined
+      : clientEncoderForDelivery(options, logCtx, !!routedCompaction, transportState.activeAdapter.name);
     if (clientEncoder) {
       return deliverClientEncodedResponse({
         encoder: clientEncoder,
@@ -184,9 +206,10 @@ export async function deliverAdapterResponse(
         declaredToolNames,
         bareCustomToolNames,
         enforceDeclaredToolNames: options.inboundWire !== "chat" && options.inboundWire !== "anthropic",
+        terminalRefusalAsFailed: options.inboundWire !== "chat" && options.inboundWire !== "anthropic",
         toolParameterSchemas,
         ...(options.onFirstOutput ? { onFirstOutput: options.onFirstOutput } : {}),
-        ...(routedCompaction ? { compaction: true } : {}),
+        ...(routedCompaction ? { compaction: true, compactionRetention: parsed._compactionRetention } : {}),
         // Same grok-surface split as the runTurn branch above.
         ...(logCtx.surface === "grok" ? { heartbeatStyle: "comment" as const } : {}),
         onUsage: usage => {
@@ -240,6 +263,17 @@ export async function deliverAdapterResponse(
       } else {
         guardedEvents = initialEvents;
       }
+      // Optional advisor guard — same semantics as the streaming branch above.
+      if (parsed._advisorGuard) {
+        const advisorEvents: AdapterEvent[] = [];
+        for await (const event of parsed._advisorGuard({
+          parsed,
+          translatorBudget,
+          firstEvents: (async function* () { yield* guardedEvents; })(),
+          continuation: fetchTerminalGuardContinuation,
+        })) advisorEvents.push(event);
+        guardedEvents = advisorEvents;
+      }
       if (emptyCompletionGuardEnabled) {
         events = [];
         for await (const event of guardEmptyCompletionEventStream({
@@ -257,8 +291,9 @@ export async function deliverAdapterResponse(
     } finally {
       cleanupUpstreamAbort();
     }
-    const { toolNsMap, declaredToolNames, toolParameterSchemas, freeformToolNames, bareCustomToolNames, toolSearchToolNames } = toolBridgeMaps;
+    const { toolNsMap, declaredToolNames, toolParameterSchemas, freeformToolNames, bareCustomToolNames, toolSearchToolNames } = deliveryToolBridgeMaps;
     let providerState: OcxProviderContinuationState | undefined;
+    events = guardEmptyToolCatalogEvents(events, clientToolCatalog, translatorBudget);
     const json = buildResponseJSON(events, parsed._responseModelId ?? parsed.modelId, {
       translatorBudget,
       replayCacheScope: parsed._reasoningReplayScope,
@@ -271,7 +306,7 @@ export async function deliverAdapterResponse(
       toolParameterSchemas,
       freeformToolNames,
       toolSearchToolNames,
-      ...(routedCompaction ? { compaction: true } : {}),
+      ...(routedCompaction ? { compaction: true, compactionRetention: parsed._compactionRetention } : {}),
       onProviderState: state => { providerState = state; },
       onUsage: usage => {
         transportState.bindKeyUsageFromBridge(usage);

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

@@ -54,7 +54,7 @@ import { describeUpstreamConnectFailure } from "./upstream-error";
 import type { OpaqueBlobRecoveryGuard } from "./core-opaque-recovery";
 import type { AttemptRecoveryKind } from "../../usage/log";
 import type { OAuthAccessSnapshot } from "../../oauth";
-import { OAuthAccountPausedError, OAuthLoginRequiredError, publicOAuthAuthenticationErrorMessage } from "../../oauth";
+import { OAuthAccountPausedError, OAuthLoginRequiredError, OAuthTokenRefreshBusyError, publicOAuthAuthenticationErrorMessage } from "../../oauth";
 import { getAccountSet, markAccountNeedsReauthIfGeneration } from "../../oauth/store";
 import { tryKiroAlternateAfterTerminalRefresh } from "../../oauth/kiro-terminal-failover";
 import { classifyKiroRefusal } from "../../adapters/kiro-refusal";
@@ -112,6 +112,7 @@ import { mapCodexAuthContextErrorToResponse } from "./codex-auth-error";
 import { chargeWorkflowSends } from "../../lib/workflow-budget";
 import { isAntigravityValidationRefusal } from "./antigravity-validation-refusal";
 import { unboundPoolSpendRefusalResponse } from "../workflow-refusal";
+import { AdmissionModelDeniedError, admissionModelDeniedResponse } from "../admission-model-scope";
 
 /** One responsibility of the Responses request pipeline; state owners are explicit. */
 export async function prepareAdapterExchange(
@@ -225,8 +226,13 @@ export async function prepareAdapterExchange(
 
 
   let selfContainedJudgment: boolean | undefined;
+  // On the Antigravity direct lane the raw body carries only settings; judge the body the bridge
+  // would have sent, built only if an ambiguous reset actually asks.
+  const directRequest = options.directRequest;
   const requestIsSelfContained = (): boolean =>
-    selfContainedJudgment ??= selfContainedResponsesBody(parsed._rawBody);
+    selfContainedJudgment ??= selfContainedResponsesBody(directRequest
+      ? { ...(parsed._rawBody as Record<string, unknown>), ...directRequest.bridgeConversation() }
+      : parsed._rawBody);
   const claimPreHeaderResend = (): boolean => authorizeResendForRecovery(
     "pre-header", "connection-reset",
     ambiguousResendAllowanceFor(route.provider, requestIsSelfContained, claimAmbiguousResend),
@@ -541,6 +547,10 @@ export async function prepareAdapterExchange(
       return storageRefusal;
     }
     const refusal = err instanceof UpstreamRetryEvidenceError ? err.cause : err;
+    if (refusal instanceof AdmissionModelDeniedError) {
+      releaseCodexAuthContextProbeLease(admissionState.authCtx);
+      return admissionModelDeniedResponse(refusal);
+    }
     const codexRefusal = mapCodexAuthContextErrorToResponse(unwrapUpstreamRetryEvidenceError(err), {
       now: Date.now(), accountSelector: route.codexAccountNamespace,
     });
@@ -810,6 +820,13 @@ export async function prepareAdapterExchange(
           releaseCodexAuthContextProbeLease(admissionState.authCtx);
           return { failed: storageRefusal };
         }
+        const scopeRefusal = unwrapUpstreamRetryEvidenceError(err);
+        if (!options.abortSignal?.aborted && scopeRefusal instanceof AdmissionModelDeniedError) {
+          cleanupUpstreamAbort();
+          upstream.abort();
+          releaseCodexAuthContextProbeLease(admissionState.authCtx);
+          return { failed: admissionModelDeniedResponse(scopeRefusal) };
+        }
         const codexRefusal = !options.abortSignal?.aborted && mapCodexAuthContextErrorToResponse(unwrapUpstreamRetryEvidenceError(err), {
           now: Date.now(), accountSelector: route.codexAccountNamespace,
         });
@@ -915,6 +932,15 @@ export async function prepareAdapterExchange(
         try {
           refreshed = await refreshResolvedOAuthSelection(transportState.sentOAuthSnapshot);
         } catch (err) {
+          const cancelled = options.abortSignal?.aborted || req.signal.aborted;
+          if (cancelled || err instanceof OAuthTokenRefreshBusyError) {
+            cleanupUpstreamAbort();
+            upstream.abort();
+            try { void upstreamResponse.body?.cancel().catch(() => {}); } catch { /* already closed */ }
+            releaseCodexAuthContextProbeLease(admissionState.authCtx);
+            return cancelled ? clientCancelledResponse()
+              : formatErrorResponse(503, "server_error", publicOAuthAuthenticationErrorMessage(err), { retryAfter: "1" });
+          }
           const failed = transportState.sentOAuthSnapshot;
           if (route.providerName === "google-antigravity" && !antigravityAuthRotationAttempted
             && err instanceof OAuthLoginRequiredError && failed
@@ -1320,9 +1346,17 @@ export async function prepareAdapterExchange(
             resolveWireProtocolOverride(route.providerName, route.modelId, route.provider, inboundWire, route.staticPolicy),
             config.cacheRetention,
           );
+          bindRouteReasoningReplayScope({
+            parsed,
+            providerName: route.providerName,
+            provider: route.provider,
+            adapterName: transportState.activeAdapter.name,
+            oauthCredentialSnapshot: transportState.replayOAuthCredentialSnapshot,
+          });
           sealRequestAttemptIdentity(logCtx.activeAttempt, logCtx.provider, transportState.activeAdapter.name, logCtx.accountLogLabel);
           recordAttemptCredentialSource(logCtx.activeAttempt, route.providerName, route.provider, transportState.activeAdapter.name);
-         const result = await rebuildAndRefetch(upstreamResponse.status === 401 ? "oauth-401" : "anthropic-oauth-429");
+         const result = await rebuildAndRefetch(upstreamResponse.status === 401 ? "oauth-401"
+           : upstreamResponse.status === 403 ? "oauth-account-403" : "anthropic-oauth-429");
          if ("failed" in result) return result.failed;
          upstreamResponse = result;
           if (isNonReplayableResponse(upstreamResponse)) continue recovery;
@@ -1539,7 +1573,17 @@ export async function prepareAdapterExchange(
           // The hop's permit is already settled by the dispatch boundary above; continuing
           // only skips the remaining arms, it does not abandon a reservation.
           if (isNonReplayableResponse(upstreamResponse)) continue recovery;
-       } catch {
+       } catch (err) {
+          // A local admission denial is terminal for the whole request: rotating to another
+          // account cannot re-scope the key, so the refusal answers in place of the 429.
+          const scopeRefusal = unwrapUpstreamRetryEvidenceError(err);
+          if (!options.abortSignal?.aborted && scopeRefusal instanceof AdmissionModelDeniedError) {
+            hop.permit?.release();
+            cleanupUpstreamAbort();
+            upstream.abort();
+            releaseCodexAuthContextProbeLease(admissionState.authCtx);
+            return admissionModelDeniedResponse(scopeRefusal);
+          }
          // A throw before the send — snapshot fetch, credential application, adapter
           // resolution — must hand the reservation back. Without this the ladder charges the
           // request for a send it never made, and a later recovery in the same request is

package/src/server/responses/advisor-plan-slot.ts

@@ -0,0 +1,27 @@
+/** Core-owned, config-scoped registration. An inactive install has no Advisor factory. */
+import type { OcxConfig, OcxParsedRequest, OcxTool } from "../../types";
+
+export interface AdvisorPlanInput {
+  config: OcxConfig;
+  workerIdentity: string;
+  workerModelId: string;
+  abortSignal?: AbortSignal;
+}
+
+export interface RegisteredAdvisorPlan {
+  tool: OcxTool;
+  preflightInject(parsed: OcxParsedRequest): Promise<boolean>;
+  attachGuard(parsed: OcxParsedRequest): void;
+}
+
+type AdvisorPlanFactory = (input: AdvisorPlanInput) => Promise<RegisteredAdvisorPlan | null>;
+const factories = new WeakMap<OcxConfig, AdvisorPlanFactory>();
+
+export function setAdvisorPlanFactory(config: OcxConfig, factory: AdvisorPlanFactory | null): void {
+  if (factory) factories.set(config, factory);
+  else factories.delete(config);
+}
+
+export function createRegisteredAdvisorPlan(input: AdvisorPlanInput): Promise<RegisteredAdvisorPlan | null> | null {
+  return factories.get(input.config)?.(input) ?? null;
+}

package/src/server/responses/advisor-retention.ts

@@ -0,0 +1,117 @@
+import { jsonUtf8Bytes } from "../../lib/json-byte-size";
+import {
+  TRANSLATOR_MAX_TURN_BYTES,
+  TranslatorBudgetExceededError,
+  type TranslatorBudget,
+} from "../../lib/translator-budget";
+import type { AdapterEvent, OcxMessage, OcxUsage } from "../../types";
+import { mergeUsage } from "./terminal-guard";
+
+export const MAX_ADVISOR_RETAINED_EVENTS = 4_096;
+export const MAX_ADVISOR_RETAINED_CALLS = 1_024;
+
+/** Independent ownership: never borrow the bridge's event or tool-call leases. */
+export class AdvisorRetention {
+  private legBytes = 0;
+  private historyBytes = 0;
+  private copyBytes = 0;
+  private usageBytes = 0;
+  private events = 0;
+  private calls = 0;
+
+  constructor(private readonly shared?: TranslatorBudget) {}
+
+  private charge(bytes: number, kind: "retained_collectors" | "request_copies"): void {
+    if (bytes > TRANSLATOR_MAX_TURN_BYTES - this.legBytes - this.historyBytes - this.copyBytes - this.usageBytes) {
+      throw new TranslatorBudgetExceededError(kind, TRANSLATOR_MAX_TURN_BYTES);
+    }
+    this.shared?.chargeRetained(bytes, { kind });
+  }
+
+  /** Usage can alias a terminal payload. Keep its own owner after the raw leg ends. */
+  usage(first: OcxUsage | undefined, second: OcxUsage | undefined): OcxUsage | undefined {
+    if (!second) return first;
+    // Two-input merge materializes a new object. Reserve before merging, including
+    // a conservative bound for concatenation if a provider supplied malformed scalars.
+    const bytes = first
+      ? (jsonUtf8Bytes(first) + jsonUtf8Bytes(second)) * 2 + 256
+      : jsonUtf8Bytes(second) + 1;
+    this.charge(bytes, "request_copies");
+    this.copyBytes += bytes;
+    let retained = 0;
+    try {
+      const merged = mergeUsage(first, second);
+      const measured = jsonUtf8Bytes(merged, bytes) + 1;
+      if (measured > bytes) throw new TranslatorBudgetExceededError("request_copies", TRANSLATOR_MAX_TURN_BYTES);
+      retained = measured;
+      const previous = this.usageBytes;
+      this.usageBytes = retained;
+      this.shared?.releaseRetained(previous, { kind: "request_copies" });
+      return merged;
+    } finally {
+      this.copyBytes -= bytes;
+      this.shared?.releaseRetained(bytes - retained, { kind: "request_copies" });
+    }
+  }
+
+  /**
+   * `counted: false` is only for non-empty text/thinking deltas: the leg's 16 KiB character
+   * caps already bound how many of those can be held, and an ordinary long answer streams
+   * thousands of them. They still consume bytes.
+   */
+  retain(event: AdapterEvent, call = false, counted = true): void {
+    if ((counted && this.events >= MAX_ADVISOR_RETAINED_EVENTS) || (call && this.calls >= MAX_ADVISOR_RETAINED_CALLS)) {
+      throw new TranslatorBudgetExceededError("retained_collectors", TRANSLATOR_MAX_TURN_BYTES);
+    }
+    const bytes = jsonUtf8Bytes(event) + 1;
+    this.charge(bytes, "retained_collectors");
+    this.legBytes += bytes;
+    if (counted) this.events++;
+    if (call) this.calls++;
+  }
+
+  /** Hold a bounded materialization while its raw sources are still retained. */
+  copy<T>(bytes: number, build: () => T): T {
+    this.charge(bytes, "request_copies");
+    this.copyBytes += bytes;
+    try { return build(); }
+    finally {
+      this.copyBytes -= bytes;
+      this.shared?.releaseRetained(bytes, { kind: "request_copies" });
+    }
+  }
+
+  /** Reserve before constructing replay JSON; transfer the reservation to history. */
+  history<T extends OcxMessage | undefined>(bytes: number, build: () => T): T {
+    this.charge(bytes, "request_copies");
+    this.copyBytes += bytes;
+    let retained = 0;
+    try {
+      const message = build();
+      const measured = message ? jsonUtf8Bytes(message, bytes) + 1 : 0;
+      if (measured > bytes) throw new TranslatorBudgetExceededError("request_copies", TRANSLATOR_MAX_TURN_BYTES);
+      retained = measured;
+      this.historyBytes += retained;
+      return message;
+    } finally {
+      this.copyBytes -= bytes;
+      this.shared?.releaseRetained(bytes - retained, { kind: "request_copies" });
+    }
+  }
+
+  /** Event envelopes dominate the replay content; allow new message/call framing too. */
+  replayReservation(): number { return this.legBytes + (this.calls + 1) * 256; }
+
+  releaseLeg(): void {
+    this.shared?.releaseRetained(this.legBytes, { kind: "retained_collectors" });
+    this.legBytes = this.events = this.calls = 0;
+  }
+
+  release(): void {
+    this.releaseLeg();
+    this.shared?.releaseRetained(this.historyBytes, { kind: "request_copies" });
+    this.historyBytes = 0;
+    this.shared?.releaseRetained(this.usageBytes, { kind: "request_copies" });
+    this.usageBytes = 0;
+  }
+}