Merge remote-tracking branch 'origin/release/v3.8.50' into HEAD

Co-authored-by: diegosouzapw <diegosouzapw@users.noreply.github.com>

changelog.d/features/9544-muse-code-cli-provider.md

@@ -0,0 +1 @@
+- feat(providers): add Muse Code CLI provider preset (#9544)

open-sse/config/providers/index.ts

@@ -225,6 +225,7 @@ import { digitaloceanProvider } from "./registry/digitalocean/index.ts";
 import { hcnsecProvider } from "./registry/hcnsec/index.ts";
 import { promptqlProvider } from "./registry/promptql/index.ts";
 import { hyperagentProvider } from "./registry/hyperagent/index.ts";
+import { muse_codeProvider } from "./registry/muse-code/index.ts";
 
 export const REGISTRY: Record<string, RegistryEntry> = {
   aimlapi: aimlapiProvider,
@@ -451,5 +452,6 @@ export const REGISTRY: Record<string, RegistryEntry> = {
   hcnsec: hcnsecProvider,
   promptql: promptqlProvider,
   hyperagent: hyperagentProvider,
+  "muse-code": muse_codeProvider,
   unorouter: unorouterProvider,
 };

open-sse/config/providers/registry/muse-code/index.ts

@@ -0,0 +1,106 @@
+import type { RegistryEntry } from "../../shared.ts";
+import { buildOpenAiCompatibleRegistryEntry } from "../../shared.ts";
+
+/**
+ * Muse Code CLI — Meta's agentic coding tool.
+ *
+ * Wire format: OpenAI Responses API (POST /responses).
+ * Auth: Bearer token from META_API_KEY env var.
+ * Reasoning efforts: xhigh/ultra -> high (handled generically).
+ *
+ * @see https://github.com/joymadhu49/muse-openrouter-shim
+ */
+export const muse_codeProvider: RegistryEntry = buildOpenAiCompatibleRegistryEntry({
+  id: "muse-code",
+  alias: "mc",
+  passthroughModels: true,
+  defaultContextLength: 200000,
+  models: [
+    {
+      id: "llama-4-maverick",
+      name: "Llama 4 Maverick",
+      contextLength: 1048576,
+      maxOutputTokens: 131072,
+      supportsReasoning: true,
+      supportsXHighEffort: true,
+      toolCalling: true,
+      supportsVision: true,
+      targetFormat: "openai-responses",
+      unsupportedParams: ["logprobs", "topLogprobs", "logitBias"],
+    },
+    {
+      id: "llama-4-scout",
+      name: "Llama 4 Scout",
+      contextLength: 1048576,
+      maxOutputTokens: 131072,
+      supportsReasoning: true,
+      supportsXHighEffort: true,
+      toolCalling: true,
+      supportsVision: true,
+      targetFormat: "openai-responses",
+      unsupportedParams: ["logprobs", "topLogprobs", "logitBias"],
+    },
+    {
+      id: "llama-3.3-70b",
+      name: "Llama 3.3 70B",
+      contextLength: 131072,
+      maxOutputTokens: 32768,
+      supportsReasoning: false,
+      toolCalling: true,
+      targetFormat: "openai-responses",
+      unsupportedParams: ["logprobs", "topLogprobs"],
+    },
+    {
+      id: "llama-3.1-405b",
+      name: "Llama 3.1 405B",
+      contextLength: 131072,
+      maxOutputTokens: 32768,
+      supportsReasoning: false,
+      toolCalling: true,
+      targetFormat: "openai-responses",
+      unsupportedParams: ["logprobs", "topLogprobs"],
+    },
+    {
+      id: "llama-3.1-70b",
+      name: "Llama 3.1 70B",
+      contextLength: 131072,
+      maxOutputTokens: 32768,
+      supportsReasoning: false,
+      toolCalling: true,
+      targetFormat: "openai-responses",
+      unsupportedParams: ["logprobs", "topLogprobs"],
+    },
+    {
+      id: "llama-3.1-8b",
+      name: "Llama 3.1 8B",
+      contextLength: 131072,
+      maxOutputTokens: 32768,
+      supportsReasoning: false,
+      toolCalling: true,
+      targetFormat: "openai-responses",
+      unsupportedParams: ["logprobs", "topLogprobs"],
+    },
+    {
+      id: "llama-3.2-90b-vision",
+      name: "Llama 3.2 90B Vision",
+      contextLength: 131072,
+      maxOutputTokens: 32768,
+      supportsReasoning: false,
+      toolCalling: true,
+      supportsVision: true,
+      targetFormat: "openai-responses",
+      unsupportedParams: ["logprobs", "topLogprobs"],
+    },
+    {
+      id: "llama-3.2-11b-vision",
+      name: "Llama 3.2 11B Vision",
+      contextLength: 131072,
+      maxOutputTokens: 32768,
+      supportsReasoning: false,
+      toolCalling: true,
+      supportsVision: true,
+      targetFormat: "openai-responses",
+      unsupportedParams: ["logprobs", "topLogprobs"],
+    },
+  ],
+});

open-sse/services/combo/validateQuality.ts

@@ -190,10 +190,26 @@ function isRecord(value: unknown): value is Record<string, unknown> {
   return !!value && typeof value === "object" && !Array.isArray(value);
 }
 
+/**
+ * Whether an `error` field carries a real failure signal. A key-presence check
+ * (`!= null`) false-positives on benign values some backends emit on every
+ * chunk (`{}`, `""`, `false`, `0`) — e.g. tool-call turns where a chunk with
+ * real tool_calls content also carries `"error": {}`. Only substantive values
+ * are treated as upstream failures.
+ */
+function isSubstantiveError(value: unknown): boolean {
+  if (value === null || value === undefined) return false;
+  if (typeof value === "string") return value.trim().length > 0;
+  if (typeof value === "object" && !Array.isArray(value)) {
+    return Object.keys(value as Record<string, unknown>).length > 0;
+  }
+  return value === true;
+}
+
 function isStreamingUpstreamError(parsed: unknown, eventType: string): boolean {
   if (eventType === "response.failed" || eventType === "error") return true;
   if (!isRecord(parsed)) return false;
-  if (parsed.error != null) return true;
+  if (isSubstantiveError(parsed.error)) return true;
 
   const nestedResponse = isRecord(parsed.response) ? parsed.response : null;
   return nestedResponse?.status === "failed" && nestedResponse.error != null;

scripts/quality/validate-release-green.mjs

@@ -609,14 +609,6 @@ async function main() {
         args: ["run", "check:pack-artifact"],
         timeout: 20 * 60 * 1000,
       });
-      // WS1.2 (#7065 class): boot the REAL packed tarball from a clean install —
-      // the runtime gate structure checks cannot provide. Reuses the same dist/ build.
-      slow.push({
-        id: "pack-boot",
-        label: "Tarball boot-smoke (installed CLI serves /health)",
-        args: ["run", "check:pack-boot"],
-        timeout: 15 * 60 * 1000,
-      });
     }
     slow.forEach((g) => announce(`${g.label} [parallel]`));
     const slowResults = await Promise.all(
@@ -633,6 +625,41 @@ async function main() {
         detail: code === 0 ? "pass" : firstFailureLine(out),
       });
     });
+
+    if (WITH_BUILD) {
+      // WS1.2 (#7065 class): boot the REAL packed tarball from a clean install.
+      // check:pack-artifact is the builder for dist/ when staging is absent, so the
+      // boot smoke MUST run after it completes. Running both in the parallel wave
+      // races check:pack-boot against dist/server.js creation on clean worktrees.
+      const packArtifactIndex = slow.findIndex((g) => g.id === "pack-artifact");
+      const packArtifactResult = slowResults[packArtifactIndex];
+      const bootLabel = "Tarball boot-smoke (installed CLI serves /health)";
+
+      if (!packArtifactResult || packArtifactResult.code !== 0) {
+        const out = "skipped because package-artifact did not produce a valid dist/ build";
+        saveGateLog("pack-boot", out);
+        record({
+          id: "pack-boot",
+          label: bootLabel,
+          kind: "hard",
+          ok: false,
+          detail: out,
+        });
+      } else {
+        announce(bootLabel);
+        const { code, out } = await runAsync(npmCmd, ["run", "check:pack-boot"], {
+          timeout: 15 * 60 * 1000,
+        });
+        saveGateLog("pack-boot", out);
+        record({
+          id: "pack-boot",
+          label: bootLabel,
+          kind: "hard",
+          ok: code === 0,
+          detail: code === 0 ? "pass" : firstFailureLine(out),
+        });
+      }
+    }
   } else if (WITH_BUILD) {
     // --with-build without the suites (--quick): still verify the package artifact.
     const { code, out } = await runAsync(npmCmd, ["run", "check:pack-artifact"], {

src/app/api/v1/muse-code/models/route.ts

@@ -0,0 +1,87 @@
+/**
+ * Muse Code CLI proprietary model catalog endpoint.
+ *
+ * Muse CLI calls GET /muse-code/models (or --base-url/muse-code/models)
+ * to discover available models. Returns the proprietary Muse format:
+ *
+ *   { object: "list", data: [{ id, object, created, owned_by, metadata }] }
+ *
+ * Each model's metadata includes: name, family, reasoning, tool_call,
+ * modalities, limit, cost.
+ */
+
+import { muse_codeProvider } from "@omniroute/open-sse/config/providers/registry/muse-code/index.ts";
+
+const MUSECODE_TIMESTAMP = Math.floor(Date.now() / 1000);
+
+interface MuseCodeModel {
+  id: string;
+  object: "model";
+  created: number;
+  owned_by: string;
+  metadata: {
+    name: string;
+    family: string;
+    reasoning: boolean;
+    tool_call: boolean;
+    modalities: string[];
+    limit: number;
+    cost: number;
+  };
+}
+
+function buildModelCatalog(): MuseCodeModel[] {
+  const data: MuseCodeModel[] = [];
+
+  for (const model of muse_codeProvider.models) {
+    let family = "llama";
+    if (model.id.includes("llama-4")) family = "llama-4";
+    else if (model.id.includes("llama-3.3")) family = "llama-3.3";
+    else if (model.id.includes("llama-3.2")) family = "llama-3.2";
+    else if (model.id.includes("llama-3.1")) family = "llama-3.1";
+
+    const modalities: string[] = ["text"];
+    if (model.supportsVision) modalities.push("image");
+
+    data.push({
+      id: model.id,
+      object: "model",
+      created: MUSECODE_TIMESTAMP,
+      owned_by: "meta",
+      metadata: {
+        name: model.name,
+        family,
+        reasoning: !!model.supportsReasoning,
+        tool_call: !!model.toolCalling,
+        modalities,
+        limit: model.contextLength ?? 200_000,
+        cost: model.id.includes("maverick") || model.id.includes("405b") ? 3 : 1,
+      },
+    });
+  }
+
+  return data;
+}
+
+// Cache the catalog for the lifetime of the process — model list is static.
+const CATALOG = buildModelCatalog();
+const CATALOG_PAYLOAD = JSON.stringify({ object: "list", data: CATALOG }, null, 2);
+
+export async function OPTIONS() {
+  return new Response(null, {
+    headers: {
+      "Access-Control-Allow-Methods": "GET, OPTIONS",
+      "Access-Control-Allow-Headers": "*",
+    },
+  });
+}
+
+export async function GET() {
+  return new Response(CATALOG_PAYLOAD, {
+    status: 200,
+    headers: {
+      "content-type": "application/json",
+      "cache-control": "public, max-age=3600",
+    },
+  });
+}

src/shared/constants/clientIdentityProfiles.ts

@@ -51,6 +51,13 @@ const GEMINI_CLI_PROFILE: ClientIdentityProfile = Object.freeze({
     "User-Agent": "GeminiCLI/0.1.0 (linux; x64)",
   }),
 });
+const MUSE_CLI_PROFILE: ClientIdentityProfile = Object.freeze({
+  id: "muse-cli",
+  label: "Muse Code CLI",
+  headers: Object.freeze({
+    "User-Agent": "MuseCodeCLI/0.1.0 (linux; x64)",
+  }),
+});
 
 /** Ordered so `CLIENT_IDENTITY_PROFILE_OPTIONS` renders "Default" first. */
 export const CLIENT_IDENTITY_PROFILES: Readonly<Record<string, ClientIdentityProfile>> =
@@ -59,6 +66,7 @@ export const CLIENT_IDENTITY_PROFILES: Readonly<Record<string, ClientIdentityPro
     "claude-cli": CLAUDE_CLI_PROFILE,
     "codex-cli": CODEX_CLI_PROFILE,
     "gemini-cli": GEMINI_CLI_PROFILE,
+    "muse-cli": MUSE_CLI_PROFILE,
   });
 
 export const CLIENT_IDENTITY_PROFILE_IDS: readonly string[] = Object.keys(CLIENT_IDENTITY_PROFILES);

src/shared/constants/providers/apikey/frontier-labs.ts

@@ -275,4 +275,19 @@ export const APIKEY_PROVIDERS_FRONTIER = {
       "Writer Palmyra is OpenAI-compatible at https://api.writer.com/v1. palmyra-x5 offers a 1M-token context window.",
     hasFree: false,
   },
+  "muse-code": {
+    id: "muse-code",
+    alias: "mc",
+    name: "Muse Code (Meta)",
+    icon: "auto_awesome",
+    color: "#0866FF",
+    textIcon: "MC",
+    website: "https://github.com/meta-llama/llama-stack",
+    authHint:
+      "Use your META_API_KEY env var as a Bearer token. Muse Code CLI uses the OpenAI Responses API wire format (POST /responses).",
+    apiHint:
+      "Muse Code is OpenAI-compatible. OmniRoute routes chat traffic through the Responses API and exposes the proprietary model catalog at /v1/muse-code/models.",
+    passthroughModels: true,
+    hasFree: false,
+  },
 };

tests/snapshots/provider/translate-path.json

@@ -3405,6 +3405,26 @@
       "stream": "https://api.morphllm.com/v1/chat/completions"
     }
   },
+  "muse-code": {
+    "format": "openai",
+    "headers": {
+      "apiKey": {
+        "Accept": "text/event-stream",
+        "Authorization": "Bearer <TOK>",
+        "Content-Type": "application/json"
+      },
+      "nonStream": {
+        "Authorization": "Bearer <TOK>",
+        "Content-Type": "application/json"
+      },
+      "oauth": {
+        "Accept": "text/event-stream",
+        "Authorization": "Bearer <TOK>",
+        "Content-Type": "application/json"
+      }
+    },
+    "url": {}
+  },
   "muse-spark-web": {
     "format": "openai",
     "headers": {

tests/unit/compression/session-dedup-memory-7849.test.ts

@@ -80,6 +80,8 @@ test("#7849: the two-message pathological pair stays bounded", () => {
     Date.now() - started < 4000,
     "the pathological pair must stay fast; quadratic work would take seconds"
   );
+  assert.strictEqual(result.body, body, "bounded processing must preserve the input body");
+  assert.equal(result.compressed, false, "the non-deduplicable pair must fail open");
   assert.ok(Array.isArray((result.body as { messages?: unknown[] }).messages));
 });
 

tests/unit/muse-code-models.test.ts

@@ -0,0 +1,81 @@
+/**
+ * Tests for Muse Code CLI model catalog endpoint.
+ *
+ * Verifies GET /v1/muse-code/models returns the proprietary Muse format.
+ */
+
+import test from "node:test";
+import assert from "node:assert/strict";
+
+import { muse_codeProvider } from "../../open-sse/config/providers/registry/muse-code/index.ts";
+
+// ── Model catalog shape ─────────────────────────────────────────────────────
+
+test("muse-code provider has at least one model", () => {
+  assert.ok(muse_codeProvider.models.length >= 1);
+});
+
+test("muse-code models have unique ids", () => {
+  const ids = muse_codeProvider.models.map((m) => m.id);
+  const unique = new Set(ids);
+  assert.equal(unique.size, ids.length, "model IDs must be unique");
+});
+
+test("muse-code models include llama-4-maverick", () => {
+  const ids = muse_codeProvider.models.map((m) => m.id);
+  assert.ok(ids.includes("llama-4-maverick"), "must include llama-4-maverick");
+});
+
+test("muse-code models include llama-4-scout", () => {
+  const ids = muse_codeProvider.models.map((m) => m.id);
+  assert.ok(ids.includes("llama-4-scout"), "must include llama-4-scout");
+});
+
+test("muse-code models include llama-3.3-70b", () => {
+  const ids = muse_codeProvider.models.map((m) => m.id);
+  assert.ok(ids.includes("llama-3.3-70b"), "must include llama-3.3-70b");
+});
+
+test("llama-4 models have supportsXHighEffort", () => {
+  const maverick = muse_codeProvider.models.find((m) => m.id === "llama-4-maverick");
+  assert.ok(maverick, "llama-4-maverick must exist");
+  assert.equal(maverick.supportsXHighEffort, true);
+
+  const scout = muse_codeProvider.models.find((m) => m.id === "llama-4-scout");
+  assert.ok(scout, "llama-4-scout must exist");
+  assert.equal(scout.supportsXHighEffort, true);
+});
+
+test("llama-3.3-70b does not support reasoning", () => {
+  const model = muse_codeProvider.models.find((m) => m.id === "llama-3.3-70b");
+  assert.ok(model, "llama-3.3-70b must exist");
+  assert.equal(model.supportsReasoning, false);
+});
+
+test("non-reasoning models do not declare supportsXHighEffort", () => {
+  for (const model of muse_codeProvider.models) {
+    if (!model.supportsReasoning) {
+      assert.equal(
+        model.supportsXHighEffort,
+        undefined,
+        `${model.id} is not a reasoning model but has supportsXHighEffort`
+      );
+    }
+  }
+});
+
+// ── Vision models ───────────────────────────────────────────────────────────
+
+test("vision models have supportsVision: true", () => {
+  const expectedVision = [
+    "llama-4-maverick",
+    "llama-4-scout",
+    "llama-3.2-90b-vision",
+    "llama-3.2-11b-vision",
+  ];
+  for (const model of muse_codeProvider.models) {
+    if (expectedVision.includes(model.id)) {
+      assert.equal(model.supportsVision, true, `${model.id} should have supportsVision`);
+    }
+  }
+});

tests/unit/muse-code-provider.test.ts

@@ -0,0 +1,91 @@
+/**
+ * Tests for Muse Code CLI provider registry entry.
+ *
+ * Verifies the provider entry loads correctly with expected config.
+ */
+
+import test from "node:test";
+import assert from "node:assert/strict";
+
+import { muse_codeProvider } from "../../open-sse/config/providers/registry/muse-code/index.ts";
+import { getRegistryEntry } from "../../open-sse/config/providerRegistry.ts";
+
+// ── Registry entry structure ────────────────────────────────────────────────
+
+test("muse-code provider entry has id", () => {
+  assert.equal(muse_codeProvider.id, "muse-code");
+});
+
+test("muse-code provider entry has alias", () => {
+  assert.equal(muse_codeProvider.alias, "mc");
+});
+
+test("muse-code provider uses openai format", () => {
+  assert.equal(muse_codeProvider.format, "openai");
+});
+
+test("muse-code provider uses apikey auth", () => {
+  assert.equal(muse_codeProvider.authType, "apikey");
+  assert.equal(muse_codeProvider.authHeader, "bearer");
+});
+
+test("muse-code provider has passthroughModels enabled", () => {
+  assert.equal(muse_codeProvider.passthroughModels, true);
+});
+
+// ── Model entries ───────────────────────────────────────────────────────────
+
+test("muse-code provider has curated models", () => {
+  assert.ok(muse_codeProvider.models.length > 0);
+});
+
+test("all muse-code models have contextLength", () => {
+  for (const model of muse_codeProvider.models) {
+    assert.ok(
+      typeof model.contextLength === "number" && model.contextLength > 0,
+      `${model.id} must have positive contextLength`
+    );
+  }
+});
+
+test("all muse-code models have toolCalling: true", () => {
+  for (const model of muse_codeProvider.models) {
+    assert.equal(model.toolCalling, true, `${model.id} must have toolCalling enabled`);
+  }
+});
+
+test("all muse-code models have targetFormat: openai-responses", () => {
+  for (const model of muse_codeProvider.models) {
+    assert.equal(
+      model.targetFormat,
+      "openai-responses",
+      `${model.id} must use openai-responses target format`
+    );
+  }
+});
+
+test("reasoning models have supportsXHighEffort", () => {
+  for (const model of muse_codeProvider.models) {
+    if (model.supportsReasoning) {
+      assert.equal(
+        model.supportsXHighEffort,
+        true,
+        `${model.id} is a reasoning model but missing supportsXHighEffort`
+      );
+    }
+  }
+});
+
+// ── Registry discovery ──────────────────────────────────────────────────────
+
+test("muse-code is discoverable via getRegistryEntry", () => {
+  const entry = getRegistryEntry("muse-code");
+  assert.ok(entry, "getRegistryEntry must return muse-code entry");
+  assert.equal(entry.id, "muse-code");
+});
+
+test("muse-code is discoverable via alias", () => {
+  const entry = getRegistryEntry("mc");
+  assert.ok(entry, "getRegistryEntry must find muse-code by alias mc");
+  assert.equal(entry.id, "muse-code");
+});

tests/unit/quality-validation-benign-error.test.ts

@@ -0,0 +1,167 @@
+/**
+ * TDD regression guard — quality validation false-positive on benign `error`
+ * fields in streaming SSE chunks.
+ *
+ * `isStreamingUpstreamError` treats ANY non-null `error` field as an upstream
+ * failure: `parsed.error != null` is true for `{}`, `""`, `false`, and `0`.
+ * When a client like opencode issues a tool-call turn, the upstream SSE opens
+ * with role-only frames (no recognized content) and a later chunk that carries
+ * real tool_calls content PLUS a benign empty `error` field (a field some
+ * backends emit on every chunk). The error gate runs BEFORE the content
+ * recognizers, so that single frame short-circuits to "error" → 502
+ * "streaming upstream error" — while the same combo via kilocode (different
+ * wire format) never emits the empty `error` field and works fine.
+ */
+import test from "node:test";
+import assert from "node:assert/strict";
+
+const { validateResponseQuality } = await import("../../open-sse/services/combo.ts");
+
+const encoder = new TextEncoder();
+const silentLog = { warn: () => {} };
+
+function openAiSseStream(events: string[]): ReadableStream<Uint8Array> {
+  const body = events.join("\n") + "\n";
+  return new ReadableStream<Uint8Array>({
+    start(controller) {
+      controller.enqueue(encoder.encode(body));
+      controller.close();
+    },
+  });
+}
+
+/**
+ * OpenAI-compatible tool-call stream that ALSO carries a benign empty `error`
+ * field on the tool_calls chunk. Some backends emit `"error": {}` or
+ * `"error": ""` alongside every chunk; that is not a real upstream failure.
+ * The frame must be treated as CONTENT (valid), not ERROR.
+ */
+function makeToolCallStreamWithBenignError(): Response {
+  const events = [
+    // role-only first chunk — no recognized content, widens the peek window
+    `data: ${JSON.stringify({
+      id: "chatcmpl_1",
+      object: "chat.completion.chunk",
+      created: 123,
+      model: "gpt-4o",
+      choices: [{ index: 0, delta: { role: "assistant" }, finish_reason: null }],
+    })}`,
+    "",
+    // tool_calls delta + benign empty `error` field (the bug trigger)
+    `data: ${JSON.stringify({
+      id: "chatcmpl_2",
+      object: "chat.completion.chunk",
+      created: 123,
+      model: "gpt-4o",
+      choices: [
+        {
+          index: 0,
+          delta: {
+            tool_calls: [
+              { index: 0, id: "call_1", type: "function", function: { name: "Bash", arguments: "" } },
+            ],
+          },
+          finish_reason: null,
+        },
+      ],
+      error: {},
+    })}`,
+    "",
+    `data: [DONE]`,
+    "",
+  ];
+  return new Response(openAiSseStream(events), {
+    status: 200,
+    headers: { "content-type": "text/event-stream" },
+  });
+}
+
+test("OpenAI stream with tool_calls + benign empty error:{} field is VALID (not 502)", async () => {
+  const res = makeToolCallStreamWithBenignError();
+  const out = await validateResponseQuality(res, true, silentLog);
+  assert.equal(
+    out.valid,
+    true,
+    `expected valid for tool_calls chunk with benign error:{}, got valid=false (reason: ${out.reason})`
+  );
+  assert.ok(out.clonedResponse, "clonedResponse must be present for valid streaming response");
+});
+
+test("OpenAI stream with tool_calls + benign empty error:'' field is VALID", async () => {
+  const events = [
+    `data: ${JSON.stringify({
+      id: "chatcmpl_3",
+      object: "chat.completion.chunk",
+      created: 123,
+      model: "gpt-4o",
+      choices: [{ index: 0, delta: { role: "assistant" }, finish_reason: null }],
+    })}`,
+    "",
+    `data: ${JSON.stringify({
+      id: "chatcmpl_4",
+      object: "chat.completion.chunk",
+      created: 123,
+      model: "gpt-4o",
+      choices: [
+        {
+          index: 0,
+          delta: {
+            tool_calls: [
+              { index: 0, id: "call_2", type: "function", function: { name: "Read", arguments: "" } },
+            ],
+          },
+          finish_reason: null,
+        },
+      ],
+      error: "",
+    })}`,
+    "",
+    `data: [DONE]`,
+    "",
+  ];
+  const res = new Response(openAiSseStream(events), {
+    status: 200,
+    headers: { "content-type": "text/event-stream" },
+  });
+  const out = await validateResponseQuality(res, true, silentLog);
+  assert.equal(
+    out.valid,
+    true,
+    `expected valid for tool_calls chunk with benign error:"", got valid=false (reason: ${out.reason})`
+  );
+});
+
+test("Stream with a REAL non-empty error object is still flagged as invalid", async () => {
+  const events = [
+    `data: ${JSON.stringify({
+      id: "chatcmpl_5",
+      object: "chat.completion.chunk",
+      created: 123,
+      model: "gpt-4o",
+      choices: [{ index: 0, delta: { role: "assistant" }, finish_reason: null }],
+    })}`,
+    "",
+    `data: ${JSON.stringify({
+      id: "chatcmpl_6",
+      object: "chat.completion.chunk",
+      created: 123,
+      model: "gpt-4o",
+      choices: [{ index: 0, delta: {}, finish_reason: null }],
+      error: { message: "upstream quota exceeded", code: "rate_limit_exceeded" },
+    })}`,
+    "",
+    `data: [DONE]`,
+    "",
+  ];
+  const res = new Response(openAiSseStream(events), {
+    status: 200,
+    headers: { "content-type": "text/event-stream" },
+  });
+  const out = await validateResponseQuality(res, true, silentLog);
+  assert.equal(
+    out.valid,
+    false,
+    `expected invalid for real error object, got valid=true (reason: ${out.reason})`
+  );
+  assert.match(out.reason ?? "", /streaming upstream error/, "reason should mention the upstream error");
+});

tests/unit/rate-limit-queue-timeout-message-4165.test.ts

@@ -26,6 +26,12 @@ function wait(ms: number) {
   return new Promise((resolve) => setTimeout(resolve, ms));
 }
 
+// Leave enough scheduling headroom for a loaded CI/devbox while keeping the
+// executing callback longer than the queue-only budget. The actual queued-job
+// case stays short because it controls dispatch deterministically.
+const DISPATCHED_QUEUE_BUDGET_MS = 2_000;
+const QUEUED_QUEUE_BUDGET_MS = 250;
+
 test.afterEach(async () => {
   await rateLimitManager.__resetRateLimitManagerForTests();
 });
@@ -43,14 +49,23 @@ async function triggerQueueTimeout() {
     concurrentRequests: 1,
     requestsPerMinute: 100000,
     minTimeBetweenRequestsMs: 0,
-    maxWaitMs: 40,
+    maxWaitMs: DISPATCHED_QUEUE_BUDGET_MS,
   });
-  rateLimitManager.enableRateLimitProtection("conn-queue-timeout");
+  const connectionId = "conn-dispatched-timeout";
+  rateLimitManager.enableRateLimitProtection(connectionId);
 
-  return rateLimitManager.withRateLimit("openai", "conn-queue-timeout", "gpt-4o", async () => {
-    await wait(400); // > maxWaitMs (40ms) → Bottleneck fails the job
-    return "should-not-reach";
-  });
+  let dispatched = false;
+  const result = await rateLimitManager.withRateLimit(
+    "test-provider",
+    connectionId,
+    null,
+    async () => {
+      dispatched = true;
+      await wait(DISPATCHED_QUEUE_BUDGET_MS + 250);
+      return "should-not-reach";
+    }
+  );
+  return { dispatched, result };
 }
 
 async function triggerQueuedTimeout() {
@@ -60,7 +75,7 @@ async function triggerQueuedTimeout() {
     concurrentRequests: 1,
     requestsPerMinute: 0,
     minTimeBetweenRequestsMs: 0,
-    maxWaitMs: 40,
+    maxWaitMs: QUEUED_QUEUE_BUDGET_MS,
   });
   const connectionId = "conn-queued-timeout";
   rateLimitManager.enableRateLimitProtection(connectionId);
@@ -79,13 +94,12 @@ async function triggerQueuedTimeout() {
   await firstExecuting;
 
   let caught: unknown;
+  let queuedDispatched = false;
   try {
-    await rateLimitManager.withRateLimit(
-      "test-provider",
-      connectionId,
-      null,
-      async () => "should-not-dispatch"
-    );
+    await rateLimitManager.withRateLimit("test-provider", connectionId, null, async () => {
+      queuedDispatched = true;
+      return "should-not-dispatch";
+    });
     assert.fail("expected the queued job to expire");
   } catch (error) {
     caught = error;
@@ -93,16 +107,20 @@ async function triggerQueuedTimeout() {
     releaseFirst();
     await first;
   }
-  return caught;
+  return { caught, queuedDispatched };
 }
 
 test("#4165 a dispatched provider call is not killed by the queue budget", async () => {
-  const result = await triggerQueueTimeout();
-  assert.equal(result, "should-not-reach");
+  const execution = await triggerQueueTimeout();
+  assert.equal(execution.dispatched, true, "the callback must enter execution");
+  assert.equal(execution.result, "should-not-reach");
 });
 
 test("#4165 queue expiry surfaces a clear local error", async () => {
-  const caught = (await triggerQueuedTimeout()) as Error & { code?: string };
+  const result = await triggerQueuedTimeout();
+  assert.ok(result.caught instanceof Error, "queue expiry must reject with an Error");
+  assert.equal(result.queuedDispatched, false, "an expired queued callback must never dispatch");
+  const caught = result.caught as Error & { code?: string };
   assert.equal(caught.code, "RATE_LIMIT_QUEUE_TIMEOUT");
   assert.match(caught.message, /maxWaitMs/);
   assert.match(caught.message, /not an upstream/i);

tests/unit/search-route.test.ts

@@ -452,7 +452,13 @@ test("v1 search POST returns 400 when auto-select finds no configured provider (
     const body = (await response.json()) as any;
 
     assert.equal(response.status, 400);
+    assert.equal(capturedUrl, "", "fallback-only SearXNG must not receive an upstream request");
     assert.ok(body.error?.message || body.error);
+    assert.match(
+      String(body.error?.message ?? body.error),
+      /provider|configured/i,
+      "the response must explain that no provider was selected"
+    );
   } finally {
     globalThis.fetch = originalFetch;
   }

tests/unit/translator-claude-to-gemini.test.ts

@@ -200,6 +200,16 @@ test("Claude -> Gemini omits unsigned functionCall instead of injecting a fake t
     false,
     "signature-less tool_use must not become a native functionCall"
   );
+  assert.equal(
+    JSON.stringify(result).includes('"thoughtSignature"'),
+    false,
+    "the translator must not synthesize a fake thought signature"
+  );
+  assert.equal(
+    JSON.stringify(result).includes("read_file"),
+    false,
+    "the omitted unsigned call must not leak its tool payload elsewhere"
+  );
 });
 
 test("Claude -> Gemini sanitizes long tool names and exposes a restore map", () => {

tests/unit/validate-release-green.test.ts

@@ -22,16 +22,12 @@ const extract = extractCiGates as (
 ) => { id: string; job: string; args: string[]; env?: Record<string, string> }[];
 
 test("eslintCounts sums errors + warnings across files", () => {
-  const parsed = [
-    { errorCount: 2, warningCount: 5 },
-    { errorCount: 0, warningCount: 3 },
-    {},
-  ];
+  const parsed = [{ errorCount: 2, warningCount: 5 }, { errorCount: 0, warningCount: 3 }, {}];
   assert.deepEqual(eslintCounts(parsed), { errors: 2, warnings: 8 });
 });
 
 test("parseEslintJson tolerates a leading non-JSON banner", () => {
-  const out = "npm warn something\n[{\"errorCount\":0,\"warningCount\":1}]";
+  const out = 'npm warn something\n[{"errorCount":0,"warningCount":1}]';
   assert.deepEqual(parseEslintJson(out), [{ errorCount: 0, warningCount: 1 }]);
   assert.equal(parseEslintJson("no json here"), null);
 });
@@ -52,8 +48,14 @@ test("parseEslintJson tolerates ESLint's trailing unpruned-suppressions stderr s
 });
 
 test("parseCognitiveCount reads the gate's count (en + pt)", () => {
-  assert.equal(parseCognitiveCount("[cognitive-complexity] 797 function(s) exceed the threshold (15)."), 797);
-  assert.equal(parseCognitiveCount("[cognitive-complexity] REGRESSÃO — 801 violações > baseline 797"), 801);
+  assert.equal(
+    parseCognitiveCount("[cognitive-complexity] 797 function(s) exceed the threshold (15)."),
+    797
+  );
+  assert.equal(
+    parseCognitiveCount("[cognitive-complexity] REGRESSÃO — 801 violações > baseline 797"),
+    801
+  );
   assert.equal(parseCognitiveCount("no number"), null);
 });
 
@@ -175,8 +177,16 @@ test("pre-flight wires the test-masking PR-context gate against origin/main (v3.
   );
   // run() must honor a per-gate env override so GITHUB_BASE_REF actually reaches the child
   // (routed through buildGateEnv since the --hermetic scrub was added).
-  assert.match(src, /env:\s*buildGateEnv\(opts\.env\)/, "run() must merge opts.env into the child env");
-  assert.match(src, /\.\.\.\(extra \|\| \{\}\)/, "buildGateEnv must spread the per-gate env override");
+  assert.match(
+    src,
+    /env:\s*buildGateEnv\(opts\.env\)/,
+    "run() must merge opts.env into the child env"
+  );
+  assert.match(
+    src,
+    /\.\.\.\(extra \|\| \{\}\)/,
+    "buildGateEnv must spread the per-gate env override"
+  );
 });
 
 test("pre-flight --hermetic scrubs the live-test trigger vars (2026-07-05 false-positive fix)", async () => {
@@ -214,6 +224,27 @@ test("pre-flight runs the slow suites CONCURRENTLY (v3.8.45 perf — was ~1h ser
   assert.match(src, /slow\.forEach\([\s\S]*?saveGateLog\(g\.id/, "each slow gate persists its log");
 });
 
+test("pre-flight runs tarball boot only after the package artifact builder completes", async () => {
+  const fs = await import("node:fs");
+  const src = fs.readFileSync(
+    new URL("../../scripts/quality/validate-release-green.mjs", import.meta.url),
+    "utf8"
+  );
+  const parallelWave = src.indexOf("const slowResults = await Promise.all");
+  const packBoot = src.indexOf('id: "pack-boot"');
+
+  assert.ok(parallelWave >= 0, "the parallel slow-gate wave must exist");
+  assert.ok(
+    packBoot > parallelWave,
+    "pack-boot must be declared after the parallel artifact build"
+  );
+  assert.match(
+    src,
+    /packArtifactResult[\s\S]*?check:pack-boot/,
+    "pack-boot must be explicitly sequenced from the package-artifact result"
+  );
+});
+
 // ─── --full-ci gate extraction (P0, v3.8.46 post-mortem) ─────────────────────
 
 const CI_FIXTURE = `
@@ -259,7 +290,11 @@ test("extractCiGates: pulls npm-run gate steps from the ci.yml gate jobs only",
   assert.ok(ids.includes("check:docs-all") && ids.includes("check:docs-symbols"), "multi-line run");
   // …and NON-gate steps + jobs outside the gate set are ignored.
   assert.ok(!ids.includes("build") && !ids.some((i) => i.startsWith("test:")), "no build/test-run");
-  assert.equal(gates.find((g) => g.job === "test-unit"), undefined, "test-unit job is not scanned");
+  assert.equal(
+    gates.find((g) => g.job === "test-unit"),
+    undefined,
+    "test-unit job is not scanned"
+  );
 });
 
 test("extractCiGates: preserves `-- <args>` so ratchet flags reach the script", () => {
@@ -272,7 +307,10 @@ test("extractCiGates: preserves `-- <args>` so ratchet flags reach the script",
 test("extractCiGates: skips the non-local gates (pr-evidence, codeql-ratchet)", () => {
   const ids = extract(CI_FIXTURE).map((g) => g.id);
   assert.ok(!ids.includes("check:pr-evidence"), "pr-evidence needs a PR body — skipped");
-  assert.ok(!ids.includes("check:codeql-ratchet"), "codeql-ratchet is a remote-main check — skipped");
+  assert.ok(
+    !ids.includes("check:codeql-ratchet"),
+    "codeql-ratchet is a remote-main check — skipped"
+  );
   assert.ok(FULL_CI_SKIP.has("check:pr-evidence") && FULL_CI_SKIP.has("check:codeql-ratchet"));
 });
 
@@ -295,10 +333,7 @@ test("extractCiGates: attaches GITHUB_BASE_REF=main env to test-masking + de-dup
 
 test("extractCiGates: the REAL ci.yml yields the base-reds that leaked in v3.8.46", async () => {
   const fs = await import("node:fs");
-  const yaml = fs.readFileSync(
-    new URL("../../.github/workflows/ci.yml", import.meta.url),
-    "utf8"
-  );
+  const yaml = fs.readFileSync(new URL("../../.github/workflows/ci.yml", import.meta.url), "utf8");
   const ids = new Set(extract(yaml).map((g) => g.id));
   // The exact gates that leaked to the v3.8.46 release PR because the pre-flight
   // never ran them — --full-ci now reproduces every one.