Merge PR #8795: fix(tests): stop ANTHROPIC_BASE_URL from failing the Anthropic suites (@Huang-404-Q)

packages/ai/test/anthropic-alignment.test.ts

@@ -32,7 +32,7 @@ import type {
 } from "@oh-my-pi/pi-ai/types";
 import { buildModel } from "@oh-my-pi/pi-catalog/build";
 import { removeSyncWithRetries } from "@oh-my-pi/pi-utils";
-import { withEnv } from "./helpers";
+import { withEnv, withOfficialAnthropicEndpoint } from "./helpers";
 
 const ANTHROPIC_MODEL_SPEC: ModelSpec<"anthropic-messages"> = {
 	id: "claude-sonnet-4-5",
@@ -152,6 +152,8 @@ function expectClaudeMetadataUserId(userId: string | undefined, expectedSessionI
 	}
 }
 
+withOfficialAnthropicEndpoint();
+
 describe("Anthropic request fingerprint alignment", () => {
 	it("maps Stainless arch values from explicit inputs", () => {
 		expect(mapStainlessArch("x64")).toBe("x64");

packages/ai/test/anthropic-cache-refresh.test.ts

@@ -3,6 +3,7 @@ import { streamSimple } from "@oh-my-pi/pi-ai";
 import type { CacheControlEphemeral, MessageCreateParams } from "@oh-my-pi/pi-ai/providers/anthropic-wire";
 import type { CacheRetention, Context, FetchImpl, Model, ProviderSessionState } from "@oh-my-pi/pi-ai/types";
 import { buildModel } from "@oh-my-pi/pi-catalog/build";
+import { withOfficialAnthropicEndpoint } from "./helpers";
 
 const CACHE_REFRESH_DELAY_MS = 5 * 60_000 - 15_000;
 const CACHE_TOKENS = 1_200;
@@ -211,6 +212,8 @@ afterEach(() => {
 	vi.restoreAllMocks();
 });
 
+withOfficialAnthropicEndpoint();
+
 describe("Anthropic prompt-cache refresh", () => {
 	it("replays max_tokens=0 once per interval and stops after three refreshes", async () => {
 		vi.useFakeTimers();

packages/ai/test/anthropic-fast-mode.test.ts

@@ -7,6 +7,7 @@ import {
 } from "@oh-my-pi/pi-ai/providers/anthropic";
 import type { Context, Model, ProviderSessionState, ServiceTier } from "@oh-my-pi/pi-ai/types";
 import { buildModel } from "@oh-my-pi/pi-catalog/build";
+import { withOfficialAnthropicEndpoint } from "./helpers";
 
 function makeAnthropicModel(id: string): Model<"anthropic-messages"> {
 	return buildModel({
@@ -52,6 +53,8 @@ function capturePayload(model: Model<"anthropic-messages">, opts: CaptureOptions
 	return promise;
 }
 
+withOfficialAnthropicEndpoint();
+
 describe("Anthropic priority service tier → speed='fast'", () => {
 	it("sets speed='fast' for Claude Opus 4.7 when serviceTier='priority'", async () => {
 		const payload = (await capturePayload(makeAnthropicModel("claude-opus-4-7"), {

packages/ai/test/anthropic-stream-envelope.test.ts

@@ -13,7 +13,7 @@ import type {
 import type { AssistantMessageEvent, Context, Model, ModelSpec, ProviderSessionState } from "@oh-my-pi/pi-ai/types";
 import { buildModel } from "@oh-my-pi/pi-catalog/build";
 import { structuredCloneJSON } from "@oh-my-pi/pi-utils";
-import { withEnv } from "./helpers";
+import { withEnv, withOfficialAnthropicEndpoint } from "./helpers";
 
 const model: Model<"anthropic-messages"> = buildModel({
 	id: "claude-sonnet-4-5",
@@ -351,6 +351,8 @@ afterEach(() => {
 	vi.restoreAllMocks();
 });
 
+withOfficialAnthropicEndpoint();
+
 describe("anthropic stream envelope handling", () => {
 	it("ignores duplicate message_start envelopes without resetting streamed text", async () => {
 		vi.spyOn(AnthropicMessages.prototype, "create").mockImplementation(

packages/ai/test/helpers/index.ts

@@ -1,3 +1,4 @@
+import { afterEach, beforeEach } from "bun:test";
 import * as os from "node:os";
 import * as path from "node:path";
 import type { Model } from "@oh-my-pi/pi-ai/types";
@@ -114,3 +115,29 @@ async function readAuthGatewayE2EStatus(): Promise<AuthGatewayE2EStatus> {
 	}
 	return { ok: true, token };
 }
+
+/**
+ * Neutralize `ANTHROPIC_BASE_URL` for the calling test file.
+ *
+ * The variable reroutes the effective Anthropic endpoint, and every
+ * "official endpoint" behavior — eager tool-input streaming, long cache
+ * retention, the Cowork TLS profile, the Claude Code session header — switches
+ * off once it points elsewhere. A contributor running a gateway, or running the
+ * suite from inside another agent, otherwise sees these tests fail on a clean
+ * checkout. Tests that exercise gateway routing set the variable explicitly
+ * with `withEnv` inside the test body, which still wins over this reset.
+ */
+export function withOfficialAnthropicEndpoint(): void {
+	let previous: string | undefined;
+	beforeEach(() => {
+		previous = Bun.env.ANTHROPIC_BASE_URL;
+		delete Bun.env.ANTHROPIC_BASE_URL;
+	});
+	afterEach(() => {
+		if (previous === undefined) {
+			delete Bun.env.ANTHROPIC_BASE_URL;
+		} else {
+			Bun.env.ANTHROPIC_BASE_URL = previous;
+		}
+	});
+}

packages/ai/test/leaked-thinking-stream.test.ts

@@ -15,6 +15,7 @@ import { getStreamingPartialJson, setStreamingPartialJson } from "@oh-my-pi/pi-a
 import { AssistantMessageEventStream } from "@oh-my-pi/pi-ai/utils/event-stream";
 import { wrapLeakedThinkingStream } from "@oh-my-pi/pi-ai/utils/leaked-thinking-stream";
 import { buildModel } from "@oh-my-pi/pi-catalog/build";
+import { withOfficialAnthropicEndpoint } from "./helpers";
 
 /** Minimal assistant message; `content`/`stopReason` overridden per event. */
 function msg(overrides: Partial<AssistantMessage> = {}): AssistantMessage {
@@ -92,6 +93,8 @@ async function nextToolSnapshot(iterator: AsyncIterator<AssistantMessageEvent>):
 	}
 }
 
+withOfficialAnthropicEndpoint();
+
 describe("wrapLeakedThinkingStream", () => {
 	async function runLeakedText(chunks: readonly string[]): Promise<{
 		events: AssistantMessageEvent[];

packages/coding-agent/test/fast-mode-scope.test.ts

@@ -10,6 +10,9 @@ import { AgentSession } from "@oh-my-pi/pi-coding-agent/session/agent-session";
 import { AuthStorage } from "@oh-my-pi/pi-coding-agent/session/auth-storage";
 import { SessionManager } from "@oh-my-pi/pi-coding-agent/session/session-manager";
 import { TempDir } from "@oh-my-pi/pi-utils";
+import { withOfficialAnthropicEndpoint } from "./helpers/anthropic-endpoint";
+
+withOfficialAnthropicEndpoint();
 
 describe("/fast targets the current model's service-tier family", () => {
 	let tempDir: TempDir;

packages/coding-agent/test/helpers/anthropic-endpoint.ts

@@ -0,0 +1,27 @@
+import { afterEach, beforeEach } from "bun:test";
+
+/**
+ * Neutralize `ANTHROPIC_BASE_URL` for the calling test file.
+ *
+ * The variable reroutes the effective Anthropic endpoint, and every
+ * "official endpoint" behavior — eager tool-input streaming, long cache
+ * retention, the Cowork TLS profile, the Claude Code session header — switches
+ * off once it points elsewhere. A contributor running a gateway, or running the
+ * suite from inside another agent, otherwise sees these tests fail on a clean
+ * checkout. Tests that exercise gateway routing set the variable explicitly
+ * with `withEnv` inside the test body, which still wins over this reset.
+ */
+export function withOfficialAnthropicEndpoint(): void {
+	let previous: string | undefined;
+	beforeEach(() => {
+		previous = Bun.env.ANTHROPIC_BASE_URL;
+		delete Bun.env.ANTHROPIC_BASE_URL;
+	});
+	afterEach(() => {
+		if (previous === undefined) {
+			delete Bun.env.ANTHROPIC_BASE_URL;
+		} else {
+			Bun.env.ANTHROPIC_BASE_URL = previous;
+		}
+	});
+}