From 42f41aa78a5a24b9e2d9925b5f8c1f3a30efed11 Mon Sep 17 00:00:00 2001
From: Eric Litman <eric@litman.org>
Date: Sat, 3 Oct 2026 14:53:22 -0400
Subject: [PATCH] test(e2e): native Messages on Sonnet now passes; drop the
 OSWE-278 known failure (MASTRA-768)

OSWE-278 is fixed. In a controlled A/B on studio2 a few minutes apart, native non-streaming
POST /v1/messages on claude-sonnet-4-6 returned 429
{"type":"error","error":{"type":"rate_limit_error","message":"Error"}} on CLIProxyAPI 7.3.17
and 200 with a normal message on v8.0.13-mob.1.

The live test now asserts 200, the trace header and a completed message. The isOswe278
predicate and the optional known-failure parameter of live() existed only for this test and
are deleted.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 e2e/lib/live.ts       | 21 ++++++-------------
 e2e/tests/live.e2e.ts | 48 ++++++++++++++-----------------------------
 2 files changed, 21 insertions(+), 48 deletions(-)

diff --git a/e2e/lib/live.ts b/e2e/lib/live.ts
index a4cd0918..6fb44f67 100644
--- a/e2e/lib/live.ts
+++ b/e2e/lib/live.ts
@@ -1,8 +1,7 @@
 import { AgentError, expect } from 'e2e';
 
-// Tiny input at the lowest effort answers in seconds; OSWE-278's 429 takes about 10.5 s (one
-// proxy retry). 60 s leaves headroom and stays under the runner's 120 s test timeout, so a
-// stall surfaces as this request's error, not as TEST_TIMEOUT.
+// Tiny input at the lowest effort answers in seconds. 60 s leaves headroom and stays under the
+// runner's 120 s test timeout, so a stall surfaces as this request's error, not as TEST_TIMEOUT.
 const BOUND_MS = 60_000;
 
 type Request = { body?: unknown; headers?: Record<string, string>; keyless?: boolean };
@@ -10,14 +9,9 @@ type Request = { body?: unknown; headers?: Record<string, string>; keyless?: boo
 /**
  * Sends one request to CLIPROXY_LIVE_URL and classifies failures that are not the product's:
  * missing inputs and a rejected key are configuration (exit 2); an unreachable proxy and a
- * provider 429 or 5xx are infrastructure (exit 3). `known` accepts a failure the test asserts
- * itself, matched on status and body.
+ * provider 429 or 5xx are infrastructure (exit 3).
  */
-export async function live(
-  path: string,
-  request: Request = {},
-  known?: (status: number, body: string) => boolean,
-) {
+export async function live(path: string, request: Request = {}) {
   const base = process.env.CLIPROXY_LIVE_URL;
   if (!base) throw new AgentError('TEST_SETUP_FAILED', 'CLIPROXY_LIVE_URL is not set');
   const key = process.env.CLIPROXY_CLIENT_KEY;
@@ -48,11 +42,8 @@ export async function live(
     throw new AgentError('AUTH_CREDENTIAL_INVALID', `${path} rejected CLIPROXY_CLIENT_KEY (401)`);
   }
   if (status === 429 || status >= 500) {
-    const text = await response.clone().text();
-    if (!known?.(status, text)) {
-      const excerpt = text.slice(0, 300);
-      throw new AgentError('ENVIRONMENT_UNAVAILABLE', `${path} answered ${status}: ${excerpt}`);
-    }
+    const excerpt = (await response.text()).slice(0, 300);
+    throw new AgentError('ENVIRONMENT_UNAVAILABLE', `${path} answered ${status}: ${excerpt}`);
   }
   return { response, signal };
 }
diff --git a/e2e/tests/live.e2e.ts b/e2e/tests/live.e2e.ts
index 6950f947..f24107e8 100644
--- a/e2e/tests/live.e2e.ts
+++ b/e2e/tests/live.e2e.ts
@@ -108,36 +108,18 @@ test('live: Responses on claude-sonnet-4-6 completes', onLive, async () => {
   expect((await response.json()).status).toBe('completed');
 });
 
-// OSWE-278's response as captured from studio2. A genuine rate limit carries a descriptive
-// message, so it stays an infrastructure failure (exit 3).
-const isOswe278 = (status: number, body: string) => {
-  try {
-    const { error } = JSON.parse(body);
-    return status === 429 && error?.type === 'rate_limit_error' && error?.message === 'Error';
-  } catch {
-    return false;
-  }
-};
-
-// Known failure: this asserts the bug. When OSWE-278 is fixed the status changes and the test
-// fails visibly; then assert 200 here instead.
-test(
-  'live: OSWE-278 known failure, native Messages on claude-sonnet-4-6 returns 429',
-  onLive,
-  async () => {
-    const { response } = await live(
-      '/v1/messages',
-      {
-        body: {
-          model: 'claude-sonnet-4-6',
-          max_tokens: 16,
-          messages: [{ role: 'user', content: say }],
-        },
-        headers: anthropic,
-      },
-      isOswe278,
-    );
-    expect(response.status).toBe(429);
-    expect((await response.json()).error).toEqual({ type: 'rate_limit_error', message: 'Error' });
-  },
-);
+test('live: native Messages on claude-sonnet-4-6 completes', onLive, async () => {
+  const { response } = await live('/v1/messages', {
+    body: {
+      model: 'claude-sonnet-4-6',
+      max_tokens: 16,
+      messages: [{ role: 'user', content: say }],
+    },
+    headers: anthropic,
+  });
+  expect(response.status).toBe(200);
+  expectTrace(response);
+  const message = await response.json();
+  expect(message.type).toBe('message');
+  expect(message.stop_reason).toBe('end_turn');
+});