npm - @bitkyc08/opencodex - Versions diffs - 2.8.2-preview.20260731 → 2.9.1 - Mend

@bitkyc08/opencodex 2.8.2-preview.20260731 → 2.9.1

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Files changed (164) hide show

package/README.md +24 -0
package/bin/ocx.mjs +32 -4
package/gui/dist/assets/index-CHwf3tTD.css +1 -0
package/gui/dist/assets/index-CuVjugeE.js +67 -0
package/gui/dist/index.html +2 -2
package/package.json +1 -1
package/src/adapters/anthropic-image-normalize.ts +114 -20
package/src/adapters/anthropic.ts +86 -5
package/src/adapters/azure.ts +3 -3
package/src/adapters/base.ts +7 -3
package/src/adapters/cursor/discovery.ts +14 -4
package/src/adapters/cursor/effort-map.ts +18 -6
package/src/adapters/cursor/framing.ts +102 -27
package/src/adapters/cursor/kv-store.ts +30 -3
package/src/adapters/cursor/live-models.ts +22 -2
package/src/adapters/cursor/live-transport.ts +245 -49
package/src/adapters/cursor/mcp-manager.ts +105 -8
package/src/adapters/cursor/native-exec-mcp.ts +5 -3
package/src/adapters/cursor/native-exec-shell.ts +296 -14
package/src/adapters/cursor/native-exec.ts +381 -33
package/src/adapters/cursor/protobuf-events.ts +28 -1
package/src/adapters/cursor/protobuf-request.ts +71 -39
package/src/adapters/cursor/request-builder.ts +2 -2
package/src/adapters/cursor/transport.ts +2 -0
package/src/adapters/cursor.ts +13 -2
package/src/adapters/google-antigravity-replay.ts +184 -17
package/src/adapters/google.ts +58 -8
package/src/adapters/kiro-thinking.ts +23 -9
package/src/adapters/kiro-tools.ts +49 -18
package/src/adapters/kiro.ts +377 -133
package/src/adapters/mimo-free.ts +36 -4
package/src/adapters/openai-chat.ts +89 -14
package/src/adapters/openai-responses.ts +130 -14
package/src/adapters/run-turn-queue.ts +7 -1
package/src/bridge.ts +466 -69
package/src/chat/outbound.ts +144 -38
package/src/claude/inbound-debug.ts +53 -8
package/src/claude/outbound.ts +224 -38
package/src/cli/agent-driven.ts +34 -1
package/src/cli/catalog-prewarm.ts +5 -2
package/src/cli/doctor.ts +12 -0
package/src/cli/export-command.ts +187 -0
package/src/cli/help.ts +11 -0
package/src/cli/index.ts +13 -3
package/src/cli/opencode.ts +36 -151
package/src/cli/star-prompt.ts +13 -4
package/src/cli/status-oauth.ts +12 -2
package/src/clients/config-export.ts +377 -0
package/src/codex/account-runtime-state.ts +19 -1
package/src/codex/account-store.ts +162 -82
package/src/codex/auth-api.ts +467 -159
package/src/codex/auth-context.ts +15 -2
package/src/codex/catalog/aggregation.ts +15 -0
package/src/codex/catalog/provider-fetch.ts +29 -0
package/src/codex/catalog/sync.ts +7 -1
package/src/codex/inject.ts +5 -5
package/src/codex/main-account-cache.ts +8 -1
package/src/codex/model-cache.ts +81 -2
package/src/codex/pool-rotation.ts +39 -0
package/src/codex/project-config-warnings.ts +12 -1
package/src/codex/quota.ts +35 -3
package/src/codex/routing.ts +46 -1
package/src/codex/shim.ts +10 -4
package/src/codex/subagent-model-fallback.ts +12 -0
package/src/codex/websocket-registry.ts +27 -0
package/src/combos/failover.ts +31 -1
package/src/combos/request.ts +9 -0
package/src/combos/resolve.ts +60 -4
package/src/combos/types.ts +12 -0
package/src/config.ts +510 -55
package/src/github/star-state.ts +13 -1
package/src/images/fulfill.ts +39 -1
package/src/images/loop.ts +52 -12
package/src/lib/admission.ts +83 -0
package/src/lib/app-owned-memory-stores.ts +173 -0
package/src/lib/app-owned-memory.ts +265 -0
package/src/lib/bun-stream-caps.ts +31 -7
package/src/lib/config-ownership.ts +33 -0
package/src/lib/crash-guard.ts +65 -5
package/src/lib/debug-log-buffer.ts +47 -6
package/src/lib/errors.ts +3 -0
package/src/lib/gcp-adc.ts +40 -2
package/src/lib/injection-debug-log.ts +26 -2
package/src/lib/sidecar-tracker.ts +5 -2
package/src/lib/sse-decoder.ts +257 -37
package/src/lib/state-store-registrations.ts +109 -0
package/src/lib/state-store-sweeper.ts +184 -0
package/src/lib/translator-budget.ts +356 -0
package/src/lib/windows-secret-acl.ts +33 -12
package/src/lib/winsw.ts +8 -1
package/src/oauth/anthropic-routing.ts +31 -7
package/src/oauth/google-antigravity.ts +2 -1
package/src/oauth/health.ts +30 -12
package/src/oauth/index.ts +127 -23
package/src/oauth/kiro.ts +10 -2
package/src/oauth/store.ts +165 -18
package/src/oauth/token-guardian.ts +43 -4
package/src/oauth/types.ts +2 -1
package/src/providers/base-url-choices.ts +10 -0
package/src/providers/derive.ts +12 -0
package/src/providers/key-failover.ts +12 -0
package/src/providers/openai-sidecar.ts +4 -1
package/src/providers/quota.ts +68 -7
package/src/providers/registry.ts +279 -3
package/src/responses/parser.ts +5 -1
package/src/responses/spill-store.ts +394 -0
package/src/responses/state.ts +520 -102
package/src/router.ts +18 -1
package/src/server/adapter-resolve.ts +20 -3
package/src/server/auth-cors.ts +121 -28
package/src/server/chat-completions.ts +57 -12
package/src/server/claude-messages.ts +85 -13
package/src/server/index.ts +239 -97
package/src/server/lifecycle.ts +155 -25
package/src/server/management/agent-settings-routes.ts +75 -32
package/src/server/management/api-key-usage.ts +167 -0
package/src/server/management/body.ts +35 -0
package/src/server/management/combo-routes.ts +5 -1
package/src/server/management/config-routes.ts +37 -12
package/src/server/management/logs-usage-routes.ts +41 -21
package/src/server/management/model-routes.ts +173 -53
package/src/server/management/oauth-account-routes.ts +115 -26
package/src/server/management/provider-routes.ts +56 -6
package/src/server/management/sidebar-routes.ts +50 -1
package/src/server/management/system-restart.ts +13 -6
package/src/server/management/system-routes.ts +15 -3
package/src/server/management/usage-summary-cache.ts +86 -0
package/src/server/management-api.ts +39 -5
package/src/server/management-auth.ts +65 -14
package/src/server/port-reclaim.ts +58 -12
package/src/server/ports.ts +2 -0
package/src/server/proxy-liveness.ts +51 -12
package/src/server/relay-eager.ts +20 -4
package/src/server/relay.ts +548 -154
package/src/server/request-decompress.ts +51 -4
package/src/server/request-log.ts +134 -15
package/src/server/responses/collaboration.ts +15 -4
package/src/server/responses/compact.ts +3 -0
package/src/server/responses/core.ts +241 -66
package/src/server/responses-image-gen-repair.ts +19 -5
package/src/server/responses-item-id-repair.ts +23 -5
package/src/server/sse-payload-rewrite.ts +71 -12
package/src/server/startup-health-cache.ts +14 -1
package/src/server/system-env.ts +8 -1
package/src/server/windows-tcp-drop.ts +15 -5
package/src/server/ws-bridge.ts +25 -0
package/src/service.ts +23 -3
package/src/storage/policy-job.ts +93 -23
package/src/storage/policy-worker.ts +6 -0
package/src/storage/restore-job.ts +62 -16
package/src/storage/restore-worker.ts +6 -0
package/src/storage/storage-mutation-coordinator.ts +36 -6
package/src/storage/worker-lifecycle.ts +181 -47
package/src/tray/windows.ts +45 -18
package/src/types.ts +34 -6
package/src/update/index.ts +24 -5
package/src/update/job.ts +598 -73
package/src/usage/log.ts +115 -17
package/src/usage/summary.ts +67 -2
package/src/vision/index.ts +112 -22
package/src/web-search/loop.ts +38 -6
package/src/web-search/progress-stream.ts +14 -3
package/gui/dist/assets/index-BHsKRFh9.css +0 -1
package/gui/dist/assets/index-GC0Vlu1Z.js +0 -67

package/gui/dist/index.html CHANGED Viewed

@@ -16,8 +16,8 @@
         } catch (e) {}
       })();
     </script>
-    <script type="module" crossorigin src="/assets/index-GC0Vlu1Z.js"></script>
-    <link rel="stylesheet" crossorigin href="/assets/index-BHsKRFh9.css">
+    <script type="module" crossorigin src="/assets/index-CuVjugeE.js"></script>
+    <link rel="stylesheet" crossorigin href="/assets/index-CHwf3tTD.css">
   </head>
   <body>
     <div id="root"></div>

package/package.json CHANGED Viewed

@@ -1,6 +1,6 @@
 {
   "name": "@bitkyc08/opencodex",
-  "version": "2.8.2-preview.20260731",
+  "version": "2.9.1",
   "description": "Universal provider proxy for OpenAI Codex & Claude Code — use any LLM with Codex CLI/App/SDK and Claude Code",
   "type": "module",
   "main": "./bin/package-main.mjs",

package/src/adapters/anthropic-image-normalize.ts CHANGED Viewed

@@ -20,6 +20,7 @@ import {
   TOTAL_IMAGE_BASE64_BUDGET,
   type ImageBlockRef,
 } from "./anthropic-image-guard";
+import { enforceAppOwnedMemoryBudget } from "../lib/app-owned-memory";
 /** One ladder position: dimension cap, JPEG quality attempts, per-image base64 cap. */
 export interface TierSpec {
@@ -98,50 +99,143 @@ type ProcessResult =
  * not entry count. Entries are immutable snapshots — demotions write NEW tier-suffixed
  * keys, never mutate stored values.
  */
-const CACHE_BYTE_CAP = 64 * MiB;
+export const IMAGE_NORMALIZE_CACHE_MAX_BYTES = 64 * MiB;
+const CACHE_MAX_ENTRIES = 4_096;
+const CACHE_MAX_ENTRY_BYTES = 20 * MiB;
 // "pass" = validated pass-through; "miss" = this position's ladder cannot meet its hard
 // cap for these bytes (skip straight to the next position — C-gate round 2, blocker 1).
-const cache = new Map<string, { data: string; mediaType: string } | "pass" | "miss">();
+type CacheValue = { data: string; mediaType: string } | "pass" | "miss";
+interface CacheEntry {
+  value: CacheValue;
+  sizeBytes: number;
+  metadataBytes: number;
+  storedAt: number;
+}
+interface NormalizeCacheLimits {
+  maxBytes: number;
+  maxEntries: number;
+  maxEntryBytes: number;
+}
+const DEFAULT_CACHE_LIMITS: NormalizeCacheLimits = {
+  maxBytes: IMAGE_NORMALIZE_CACHE_MAX_BYTES,
+  maxEntries: CACHE_MAX_ENTRIES,
+  maxEntryBytes: CACHE_MAX_ENTRY_BYTES,
+};
+const cacheEncoder = new TextEncoder();
+const cache = new Map<string, CacheEntry>();
+let cacheLimits = { ...DEFAULT_CACHE_LIMITS };
 let cacheBytes = 0;
+let cacheMetadataBytes = 0;
+let cacheSentinelEntries = 0;
 let encodeCalls = 0;
-function cachePut(key: string, value: { data: string; mediaType: string } | "pass" | "miss"): void {
-  const size = typeof value === "string" ? 0 : value.data.length;
+function cacheEntry(key: string, value: CacheValue): CacheEntry {
+  const keyBytes = cacheEncoder.encode(key).byteLength;
+  const valueBytes = typeof value === "string"
+    ? cacheEncoder.encode(value).byteLength
+    : cacheEncoder.encode(value.mediaType).byteLength + cacheEncoder.encode(value.data).byteLength;
+  const metadataBytes = keyBytes + (typeof value === "string"
+    ? cacheEncoder.encode(value).byteLength
+    : cacheEncoder.encode(value.mediaType).byteLength);
+  return { value, sizeBytes: keyBytes + valueBytes, metadataBytes, storedAt: Date.now() };
+}
+function deleteCacheEntry(key: string): number {
+  const entry = cache.get(key);
+  if (!entry) return 0;
+  cache.delete(key);
+  cacheBytes -= entry.sizeBytes;
+  cacheMetadataBytes -= entry.metadataBytes;
+  if (typeof entry.value === "string") cacheSentinelEntries--;
+  return entry.sizeBytes;
+}
+function cachePut(key: string, value: CacheValue): boolean {
+  const next = cacheEntry(key, value);
+  if (
+    next.sizeBytes > cacheLimits.maxEntryBytes
+    || next.sizeBytes > cacheLimits.maxBytes
+    || cacheLimits.maxEntries <= 0
+  ) return false;
   const existing = cache.get(key);
   if (existing !== undefined) {
-    cacheBytes -= typeof existing === "string" ? 0 : existing.data.length;
-    cache.delete(key); // re-insert refreshes recency and prevents double-count on concurrent misses
+    deleteCacheEntry(key); // re-insert refreshes recency and prevents double-count on concurrent misses
   }
-  while (cacheBytes + size > CACHE_BYTE_CAP && cache.size > 0) {
-    const oldest = cache.keys().next().value as string;
-    const evicted = cache.get(oldest);
-    cacheBytes -= typeof evicted === "string" || evicted === undefined ? 0 : evicted.data.length;
-    cache.delete(oldest);
+  while (cache.size + 1 > cacheLimits.maxEntries || cacheBytes + next.sizeBytes > cacheLimits.maxBytes) {
+    const oldest = cache.keys().next().value;
+    if (oldest === undefined || deleteCacheEntry(oldest) === 0) return false;
   }
-  cache.set(key, value);
-  cacheBytes += size;
+  cache.set(key, next);
+  cacheBytes += next.sizeBytes;
+  cacheMetadataBytes += next.metadataBytes;
+  if (typeof value === "string") cacheSentinelEntries++;
+  enforceAppOwnedMemoryBudget();
+  return true;
 }
 /** Read a cache entry, refreshing its recency (true LRU, C-gate round 1 blocker 5). */
-function cacheGet(key: string): { data: string; mediaType: string } | "pass" | "miss" | undefined {
-  const value = cache.get(key);
-  if (value !== undefined) {
+function cacheGet(key: string): CacheValue | undefined {
+  const entry = cache.get(key);
+  if (entry !== undefined) {
     cache.delete(key);
-    cache.set(key, value);
+    entry.storedAt = Date.now();
+    cache.set(key, entry);
   }
-  return value;
+  return entry?.value;
 }
 /** Test hooks: encoder-invocation counter + cache reset (no production caller). */
-export function getNormalizeStatsForTests(): { encodeCalls: number; cacheEntries: number; cacheBytes: number } {
-  return { encodeCalls, cacheEntries: cache.size, cacheBytes };
+export function getNormalizeStatsForTests(): {
+  encodeCalls: number;
+  cacheEntries: number;
+  cacheBytes: number;
+  sentinelEntries: number;
+  metadataBytes: number;
+  oldestAt: number | null;
+} {
+  return {
+    encodeCalls,
+    cacheEntries: cache.size,
+    cacheBytes,
+    sentinelEntries: cacheSentinelEntries,
+    metadataBytes: cacheMetadataBytes,
+    oldestAt: cache.values().next().value?.storedAt ?? null,
+  };
 }
 export function resetNormalizeStateForTests(): void {
   cache.clear();
   cacheBytes = 0;
+  cacheMetadataBytes = 0;
+  cacheSentinelEntries = 0;
   encodeCalls = 0;
 }
+export function setNormalizeCacheLimitsForTests(limits?: Partial<NormalizeCacheLimits>): void {
+  resetNormalizeStateForTests();
+  cacheLimits = limits ? { ...DEFAULT_CACHE_LIMITS, ...limits } : { ...DEFAULT_CACHE_LIMITS };
+}
+export function anthropicImageNormalizeRetainedStoreSnapshot(): {
+  count: number;
+  bytes: number;
+  evictableBytes: number;
+  pinnedBytes: number;
+  oldestAt: number | null;
+} {
+  return {
+    count: cache.size,
+    bytes: cacheBytes,
+    evictableBytes: cacheBytes,
+    pinnedBytes: 0,
+    oldestAt: cache.values().next().value?.storedAt ?? null,
+  };
+}
+export function evictOldestAnthropicImageNormalizeForBudget(): number {
+  const oldest = cache.keys().next().value;
+  return oldest === undefined ? 0 : deleteCacheEntry(oldest);
+}
 /** Default encoder: Bun.Image resize-to-fit + JPEG at the given quality. */
 const bunImageEncode: EncodeFn = async (input, spec, quality) => {
   const image = new Bun.Image(input);

package/src/adapters/anthropic.ts CHANGED Viewed

@@ -22,6 +22,7 @@ import { neutralizeIdentity } from "./identity";
 import { CLAUDE_CODE_HEADERS, claudeCodeSessionId } from "./client-fingerprint";
 import { buildNonOpenAIToolCatalogNudgeForTools } from "./tool-catalog-nudge";
 import { decodeServerSentEvents } from "../lib/sse-decoder";
+import { isTranslatorBudgetExceededError, retainTranslatedEventBatch, type TranslatorBudget } from "../lib/translator-budget";
 /** Map a user content part to an Anthropic content block (text or image source). */
 function toAnthropicContentPart(p: OcxContentPart): unknown {
@@ -266,6 +267,16 @@ function synthesizeToolUseId(): string {
   return `toolu_${crypto.randomUUID().replace(/-/g, "").slice(0, 24)}`;
 }
+/**
+ * A tool_use id that a client can actually echo back. `??` only catches a missing
+ * field, so an Anthropic-compatible relay that sends `""` or `"   "` produced a
+ * call whose id round-trips as blank — the next turn then cannot pair the result
+ * with its call. Treat blank as absent and synthesize (#765).
+ */
+function usableToolUseId(id: unknown): string {
+  return typeof id === "string" && id.trim() ? id : synthesizeToolUseId();
+}
 function toolUseArguments(input: unknown): string {
   if (typeof input === "string") {
     const trimmed = input.trim();
@@ -284,6 +295,23 @@ function toolUseArguments(input: unknown): string {
   return JSON.stringify(input ?? {});
 }
+/**
+ * Whether arguments assembled from a stream's `input_json_delta` fragments are usable.
+ * A tool block that sent no fragments at all is fine — that is a no-argument call. Anything
+ * else has to parse, because unlike the non-stream path the fragments have already been
+ * forwarded to the client and cannot be repaired after the fact.
+ */
+function streamedToolArgumentsParse(assembled: string): boolean {
+  const trimmed = assembled.trim();
+  if (!trimmed) return true;
+  try {
+    JSON.parse(trimmed);
+    return true;
+  } catch {
+    return false;
+  }
+}
 function anthropicKeyUsesBearer(provider: OcxProviderConfig): boolean {
   return provider.apiKeyTransport === "bearer";
 }
@@ -756,15 +784,17 @@ export function createAnthropicAdapter(provider: OcxProviderConfig, cacheRetenti
       return { url, method: "POST", headers, body: JSON.stringify(body) };
     },
-    async *parseStream(response: Response): AsyncGenerator<AdapterEvent> {
+    async *parseStream(response: Response, budget: TranslatorBudget): AsyncGenerator<AdapterEvent> {
       if (!response.body) {
         yield { type: "error", message: "No response body" };
         return;
       }
+      const budgetEncoder = new TextEncoder();
       let currentBlockType = "";
       let currentToolCallId = "";
       let currentToolCallName = "";
+      let currentToolCallJson = "";
       let pendingUsage: Record<string, number> | undefined;
       let pendingStopReason: string | undefined;
       let emittedDone = false;
@@ -779,7 +809,8 @@ export function createAnthropicAdapter(provider: OcxProviderConfig, cacheRetenti
         };
       };
-      for await (const record of decodeServerSentEvents(response.body, { includeComments: true })) {
+      try {
+      for await (const record of decodeServerSentEvents(response.body, { includeComments: true, translatorBudget: budget })) {
         if (record.kind === "comment") {
           yield { type: "heartbeat" };
           continue;
@@ -806,8 +837,10 @@ export function createAnthropicAdapter(provider: OcxProviderConfig, cacheRetenti
                 if (!block) break;
                 currentBlockType = block.type;
                 if (block.type === "tool_use") {
-                  currentToolCallId = block.id ?? synthesizeToolUseId();
+                  currentToolCallId = usableToolUseId(block.id);
                   currentToolCallName = toolNames.fromWire(block.name ?? "");
+                  currentToolCallJson = "";
+                  budget.openCall(currentToolCallId);
                   yield { type: "tool_call_start", id: currentToolCallId, name: currentToolCallName };
                 }
                 if (block.type === "redacted_thinking" && typeof block.data === "string") {
@@ -833,14 +866,43 @@ export function createAnthropicAdapter(provider: OcxProviderConfig, cacheRetenti
                   // so a stray signature on a non-thinking block can never be captured.
                   yield { type: "thinking_signature", signature: delta.signature };
                 } else if (delta.type === "input_json_delta" && typeof delta.partial_json === "string" && currentBlockType === "tool_use") {
+                  // Forwarded immediately: the bridge maps each delta to a client-visible
+                  // response.function_call_arguments.delta frame, so withholding fragments until
+                  // block close would leave a started call showing empty arguments. A copy is kept
+                  // to validate the assembled payload at content_block_stop.
+                  const previousBytes = budgetEncoder.encode(currentToolCallJson).byteLength;
+                  const nextBytes = previousBytes + budgetEncoder.encode(delta.partial_json).byteLength;
+                  const reservation = budget.reserveTransient(nextBytes, { kind: "tool_args", callId: currentToolCallId });
+                  try {
+                    currentToolCallJson += delta.partial_json;
+                    reservation.commitRetained();
+                    budget.releaseRetained(previousBytes, { kind: "tool_args", callId: currentToolCallId });
+                  } catch (error) {
+                    reservation.release();
+                    throw error;
+                  }
                   yield { type: "tool_call_delta", arguments: delta.partial_json };
                 }
                 break;
               }
               case "content_block_stop": {
                 if (currentBlockType === "tool_use") {
+                  // The non-stream path repairs an unparseable payload in toolUseArguments(); the
+                  // stream cannot, because the fragments are already downstream. Fail the turn
+                  // instead of ending a tool call whose arguments will not parse — the bridge's
+                  // terminal-error path cancels the open call (status incomplete) rather than
+                  // completing it before response.failed (#765).
+                  if (!streamedToolArgumentsParse(currentToolCallJson)) {
+                    yield {
+                      type: "error",
+                      message: "Anthropic stream sent malformed tool_use arguments (invalid JSON)",
+                    };
+                    return;
+                  }
                   yield { type: "tool_call_end" };
+                  budget.closeCall(currentToolCallId);
                   currentToolCallId = "";
+                  currentToolCallJson = "";
                 }
                 currentBlockType = "";
                 break;
@@ -863,6 +925,18 @@ export function createAnthropicAdapter(provider: OcxProviderConfig, cacheRetenti
               }
         }
       }
+      } catch (error) {
+        if (!isTranslatorBudgetExceededError(error)) throw error;
+        yield {
+          type: "error",
+          status: 502,
+          errorType: "upstream_error",
+          code: "translation_buffer_limit",
+          message: "upstream translation buffer exceeded the safe limit",
+        };
+      } finally {
+        if (currentToolCallId) budget.closeCall(currentToolCallId);
+      }
       if (!emittedDone) {
         // Fail closed on transport EOF. Compatible providers may omit message_stop after message_delta.stop_reason.
         if (pendingStopReason !== undefined) {
@@ -883,8 +957,11 @@ export function createAnthropicAdapter(provider: OcxProviderConfig, cacheRetenti
       }
     },
-    async parseResponse(response: Response): Promise<AdapterEvent[]> {
+    async parseResponse(response: Response, budget: TranslatorBudget): Promise<AdapterEvent[]> {
       const json = await response.json() as Record<string, unknown>;
+      const responseBytes = new TextEncoder().encode(JSON.stringify(json)).byteLength;
+      budget.chargeRetained(responseBytes, { kind: "retained_collectors" });
+      try {
       const events: AdapterEvent[] = [];
       const content = json.content as { type: string; text?: string; id?: string; name?: string; input?: unknown; thinking?: string; reasoning?: string; signature?: string; data?: string }[] | undefined;
       if (content) {
@@ -901,7 +978,7 @@ export function createAnthropicAdapter(provider: OcxProviderConfig, cacheRetenti
           } else if (block.type === "redacted_thinking" && typeof block.data === "string") {
             events.push({ type: "redacted_thinking", data: block.data });
           } else if (block.type === "tool_use") {
-            const id = block.id ?? synthesizeToolUseId();
+            const id = usableToolUseId(block.id);
             events.push({ type: "tool_call_start", id, name: toolNames.fromWire(block.name ?? "") });
             events.push({ type: "tool_call_delta", arguments: toolUseArguments(block.input) });
             events.push({ type: "tool_call_end" });
@@ -915,7 +992,11 @@ export function createAnthropicAdapter(provider: OcxProviderConfig, cacheRetenti
         usage: usageFromAnthropic(usage),
         ...(stopReason ? { stopReason } : {}),
       });
+      retainTranslatedEventBatch(events, budget);
       return events;
+      } finally {
+        budget.releaseRetained(responseBytes, { kind: "retained_collectors" });
+      }
     },
   };

package/src/adapters/azure.ts CHANGED Viewed

@@ -1,4 +1,4 @@
-import type { ProviderAdapter } from "./base";
+import type { IncomingMeta, ProviderAdapter } from "./base";
 import type { OcxParsedRequest, OcxProviderConfig } from "../types";
 import { createResponsesPassthroughAdapter } from "./openai-responses";
@@ -12,7 +12,7 @@ export function createAzureAdapter(provider: OcxProviderConfig): ProviderAdapter
     ...inner,
     name: "azure-openai",
-    async buildRequest(parsed: OcxParsedRequest) {
+    async buildRequest(parsed: OcxParsedRequest, incoming: IncomingMeta) {
       if (provider.authMode === "forward") {
         throw new Error("azure-openai does not support forward auth mode");
       }
@@ -20,7 +20,7 @@ export function createAzureAdapter(provider: OcxProviderConfig): ProviderAdapter
         throw new Error("azure-openai requires a non-empty apiKey");
       }
-      const request = await inner.buildRequest(parsed);
+      const request = await inner.buildRequest(parsed, incoming);
       const unresolvedPlaceholder = request.url.match(/\{[^}]*\}/)?.[0] ?? request.url.match(/[{}]/)?.[0];
       if (unresolvedPlaceholder) {
         throw new Error(`azure-openai baseUrl contains unresolved ${unresolvedPlaceholder} — set your real resource URL`);

package/src/adapters/base.ts CHANGED Viewed

@@ -1,8 +1,10 @@
 import type { AdapterEvent, OcxParsedRequest } from "../types";
+import type { TranslatorBudget } from "../lib/translator-budget";
 /** Metadata about the caller's incoming request, for auth-forwarding adapters. */
 export interface IncomingMeta {
   headers: Headers;
+  translatorBudget: TranslatorBudget;
   abortSignal?: AbortSignal;
   /**
    * Image-normalization ladder bias for upstream-413 tightened retries: every image
@@ -26,12 +28,12 @@ export interface ProviderAdapter {
    * (e.g. Vertex AI ADC token) return a Promise. Sync adapters return the object directly; callers
    * must `await` the result (awaiting a non-Promise is a no-op).
    */
-  buildRequest(parsed: OcxParsedRequest, incoming?: IncomingMeta): AdapterRequest | Promise<AdapterRequest>;
+  buildRequest(parsed: OcxParsedRequest, incoming: IncomingMeta): AdapterRequest | Promise<AdapterRequest>;
   fetchResponse?(request: AdapterRequest, ctx?: AdapterFetchContext): Promise<Response>;
-  parseStream(response: Response): AsyncGenerator<AdapterEvent>;
-  parseResponse?(response: Response): Promise<AdapterEvent[]>;
+  parseStream(response: Response, budget: TranslatorBudget): AsyncGenerator<AdapterEvent>;
+  parseResponse?(response: Response, budget: TranslatorBudget): Promise<AdapterEvent[]>;
   runTurn?(
     parsed: OcxParsedRequest,
     incoming: IncomingMeta,
@@ -44,6 +46,8 @@ export interface AdapterRequest {
     method: string;
     headers: Record<string, string>;
     body: string;
+    /** Releases observation of a serialized request body after its final fetch attempt settles. */
+    releaseBodyObservation?: () => void;
     /** Exact reasoning parameter emitted by the adapter, for request-log diagnostics only. */
     reasoningLog?: {
       effectiveEffort: string;

package/src/adapters/cursor/discovery.ts CHANGED Viewed

@@ -1,4 +1,8 @@
-import { CANONICAL_EFFORT_SUFFIXES, cursorModelEffortLadder } from "./effort-map";
+import {
+  CANONICAL_EFFORT_SUFFIXES,
+  cursorModelEffortLadder,
+  cursorWireModelIdWithEffort,
+} from "./effort-map";
 export interface CursorModelInfo {
   id: string;
@@ -66,14 +70,20 @@ function stripCursorWirePrefix(id: string): string {
 /**
  * True when a configured Cursor base model should remain exposed after live GetUsableModels filtering.
- * Live ids are full effort-suffixed variants (`claude-4.6-opus-high`); base ids match exactly or by prefix.
+ * Live ids are full effort-suffixed variants (`claude-4.6-opus-high`); base ids match exactly, the
+ * ordinary `{base}-{effort}` form, or Cursor's current `{base-without-fast}-{effort}-fast` form.
  */
 export function isCursorModelAvailableForAccount(modelId: string, liveIds: readonly string[]): boolean {
   return liveIds.some(raw => {
     const id = stripCursorWirePrefix(raw);
     if (id === modelId) return true;
-    const effortPrefix = `${modelId}-`;
-    return id.startsWith(effortPrefix) && CANONICAL_EFFORT_SUFFIXES.has(id.slice(effortPrefix.length));
+    for (const effort of CANONICAL_EFFORT_SUFFIXES) {
+      if (
+        id === `${modelId}-${effort}` ||
+        id === cursorWireModelIdWithEffort(modelId, effort)
+      ) return true;
+    }
+    return false;
   });
 }

package/src/adapters/cursor/effort-map.ts CHANGED Viewed

@@ -3,8 +3,9 @@
  *
  * Cursor model ids encode the reasoning effort as a suffix (`claude-4.6-opus-high`), and the available
  * tiers differ per model — `claude-4.6-opus` tops out at `-max`, `claude-opus-4-8` at `-xhigh`,
- * `claude-4.6-sonnet` only has `-medium`, and `composer`/`grok`/`gemini` take no suffix at all. A bare
- * id for a model that requires a suffix is rejected `ERROR_BAD_MODEL_NAME` (devlog 350.105).
+ * `claude-4.6-sonnet` only has `-medium`, and most `composer`/`gemini` models take no suffix at all.
+ * Grok Fast puts its mode marker after the effort (`grok-4.5-high-fast`). A bare id for a model that
+ * requires a suffix is rejected `ERROR_BAD_MODEL_NAME` (devlog 350.105).
  *
  * Canonical effort order is always low < medium < high < xhigh < max (max is the top tier, confirmed
  * against Anthropic docs and Cursor's live lineup). Tiers are stored in ascending canonical order.
@@ -30,10 +31,10 @@ const CURSOR_MODEL_EFFORT_TIERS: Record<string, readonly string[]> = {
   // GetUsableModels (2026-07-28) lists kimi-k3 only as effort-suffixed kimi-k3-{low,high,max};
   // the bare id returns not_found. Tiers mirror the native Kimi provider's K3 ladder.
   "kimi-k3": ["low", "high", "max"],
-  // GetUsableModels (2026-07-09) lists grok-4.5-{medium,high,xhigh} and grok-4.5-fast-{medium,high,xhigh};
-  // the bare "grok-4.5-fast" id was removed upstream and now returns not_found.
-  "grok-4.5": ["medium", "high", "xhigh"],
-  "grok-4.5-fast": ["medium", "high", "xhigh"],
+  // Cursor renamed the Grok 4.5 slugs to cursor-grok-4.5-{low,medium,high} and
+  // cursor-grok-4.5-{low,medium,high}-fast. The bare Fast id returns not_found.
+  "grok-4.5": ["low", "medium", "high"],
+  "grok-4.5-fast": ["low", "medium", "high"],
   "gpt-5.1": ["low", "high"],
   "gpt-5.1-codex-max": ["low", "medium", "high", "xhigh"],
   "gpt-5.1-codex-mini": ["low", "high"],
@@ -113,3 +114,14 @@ export function cursorModelEffortLadder(baseModelId: string): string[] | undefin
 export function cursorModelHasEffortTiers(baseModelId: string): boolean {
   return (CURSOR_MODEL_EFFORT_TIERS[baseModelId]?.length ?? 0) > 0;
 }
+/**
+ * Compose a Cursor wire id from a Codex-facing base id and effort tier.
+ * Fast variants put the mode after the effort; other models use the ordinary `{base}-{effort}` form.
+ */
+export function cursorWireModelIdWithEffort(baseModelId: string, effortSuffix: string): string {
+  if (baseModelId.endsWith("-fast")) {
+    return `${baseModelId.slice(0, -"-fast".length)}-${effortSuffix}-fast`;
+  }
+  return `${baseModelId}-${effortSuffix}`;
+}