fix(integration): restore codex and memory pipeline contracts (#4474)

Thanks @KooshaPari! Rebased onto release/v3.8.32 (clean cherry-pick — coexists with #4467's specialty changes, 118/118 unit + 12/12 memory-pipeline integration). Codex fingerprint ordering + memory-search contract fixes land.

open-sse/config/cliFingerprints.ts

@@ -49,11 +49,11 @@ export const CLI_FINGERPRINTS: Record<string, CliFingerprint> = {
       "instructions",
       "store",
       "reasoning",
+      "prompt_cache_key",
       "tools",
       "tool_choice",
       "include",
       "service_tier",
-      "prompt_cache_key",
       "client_metadata",
       "parallel_tool_calls",
       "metadata",

open-sse/mcp-server/tools/memoryTools.ts

@@ -2,7 +2,11 @@ import { z } from "zod";
 import { retrieveMemories } from "@/lib/memory/retrieval";
 import { createMemory, deleteMemory, listMemories } from "@/lib/memory/store";
 import { MemoryType } from "@/lib/memory/types";
-import { getMemorySettings, toMemoryRetrievalConfig, DEFAULT_MEMORY_SETTINGS } from "@/lib/memory/settings";
+import {
+  getMemorySettings,
+  toMemoryRetrievalConfig,
+  DEFAULT_MEMORY_SETTINGS,
+} from "@/lib/memory/settings";
 
 export const MemorySearchSchema = z.object({
   apiKeyId: z.string(),
@@ -44,7 +48,10 @@ export const memoryTools = {
 
       const config = {
         ...baseConfig,
-        maxTokens: args.maxTokens || (baseConfig.maxTokens ?? DEFAULT_MEMORY_SETTINGS.maxTokens),
+        enabled: true,
+        maxTokens:
+          args.maxTokens ??
+          (memorySettings.enabled ? memorySettings.maxTokens : DEFAULT_MEMORY_SETTINGS.maxTokens),
       };
 
       const memories = await retrieveMemories(args.apiKeyId, config);

tests/integration/memory-pipeline.test.ts

@@ -10,6 +10,7 @@ const harness = await createChatPipelineHarness("memory-pipeline");
 // The harness sets DATA_DIR before importing DB modules, so these must resolve after that.
 const { extractFactsFromText } = await import("../../src/lib/memory/extraction.ts");
 const { retrieveMemories } = await import("../../src/lib/memory/retrieval.ts");
+const { invalidateMemorySettingsCache } = await import("../../src/lib/memory/settings.ts");
 const { injectMemory, formatMemoryContext } = await import("../../src/lib/memory/injection.ts");
 const {
   BaseExecutor,
@@ -45,12 +46,14 @@ function dropFts5Artifacts() {
 test.beforeEach(async () => {
   BaseExecutor.RETRY_CONFIG.delayMs = 0;
   await resetStorage();
+  invalidateMemorySettingsCache();
   dropFts5Artifacts();
 });
 
 test.afterEach(async () => {
   BaseExecutor.RETRY_CONFIG.delayMs = harness.originalRetryDelayMs;
   await resetStorage();
+  invalidateMemorySettingsCache();
 });
 
 test.after(async () => {

tests/unit/cli-fingerprints.test.ts

@@ -0,0 +1,36 @@
+import test from "node:test";
+import assert from "node:assert/strict";
+
+const { applyFingerprint } = await import("../../open-sse/config/cliFingerprints.ts");
+
+test("Codex CLI fingerprint orders prompt_cache_key before include", () => {
+  const body = {
+    model: "gpt-5.5-low",
+    stream: true,
+    input: [{ role: "user", content: "hello" }],
+    instructions: "You are Codex.",
+    store: false,
+    reasoning: { effort: "low" },
+    tools: [],
+    tool_choice: "auto",
+    include: ["reasoning.encrypted_content"],
+    prompt_cache_key: "conv-codex",
+    service_tier: "priority",
+  };
+
+  const result = applyFingerprint("codex", {}, body);
+  const orderedKeys = Object.keys(JSON.parse(result.bodyString));
+
+  assert.deepEqual(orderedKeys.slice(0, 10), [
+    "model",
+    "stream",
+    "input",
+    "instructions",
+    "store",
+    "reasoning",
+    "prompt_cache_key",
+    "tools",
+    "tool_choice",
+    "include",
+  ]);
+});

tests/unit/memory-tools.test.ts

@@ -119,7 +119,7 @@ test("memory search respects a configured zero token budget", async () => {
   assert.equal(result.data.totalTokens, 0);
 });
 
-test("memory search keeps globally disabled memory disabled with explicit maxTokens", async () => {
+test("memory search runs explicitly even when global memory injection is disabled", async () => {
   await settingsDb.updateSettings({ memoryEnabled: false, memoryMaxTokens: 2000 });
   invalidateMemorySettingsCache();
 
@@ -137,9 +137,10 @@ test("memory search keeps globally disabled memory disabled with explicit maxTok
   });
 
   assert.equal(result.success, true);
-  assert.equal(result.data.count, 0);
-  assert.deepEqual(result.data.memories, []);
-  assert.equal(result.data.totalTokens, 0);
+  assert.equal(result.data.count, 1);
+  assert.equal(result.data.memories.length, 1);
+  assert.match(result.data.memories[0].content, /TypeScript/i);
+  assert.ok(result.data.totalTokens > 0);
 });
 
 test("memory clear deletes only older filtered entries and reports the deleted count", async () => {

tests/unit/provider-validation-specialty.test.ts

@@ -372,12 +372,12 @@ test("web-cookie provider validators surface auth and subscription failures", as
   __setPplxTlsFetchOverride(async () => {
     return { status: 403, headers: new Headers(), text: null, body: null };
   });
+  __setGrokTlsFetchOverride(async () => {
+    return { status: 401, headers: new Headers(), text: "Unauthorized", body: null };
+  });
 
   globalThis.fetch = async (url, init = {}) => {
     const target = String(url);
-    if (target.includes("grok.com/rest/app-chat/conversations/new")) {
-      return new Response(JSON.stringify({ error: "unauthorized" }), { status: 401 });
-    }
     if (target.includes("app.blackbox.ai/api/auth/session")) {
       const cookie = (init.headers as Record<string, string>)?.Cookie || "";
       if (cookie.includes("expired-cookie")) {