feat: align Go/Node DSML tool-call parsing drift tolerance and update API docs

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

tests/node/chat-stream.test.js

@@ -224,6 +224,80 @@ test('vercel stream retries thinking-only output once', async () => {
   assert.equal(parsed[2].choices[0].finish_reason, 'stop');
 });
 
+test('vercel stream switches managed account after empty retry exhaustion', async () => {
+  const originalFetch = global.fetch;
+  const fetchURLs = [];
+  const completionBodies = [];
+  const completionAuth = [];
+  let completionCalls = 0;
+  global.fetch = async (url, init = {}) => {
+    const textURL = String(url);
+    fetchURLs.push(textURL);
+    if (textURL.includes('__stream_prepare=1')) {
+      return jsonResponse({
+        session_id: 'chatcmpl-test',
+        lease_id: 'lease-test',
+        model: 'gpt-test',
+        final_prompt: 'hello',
+        thinking_enabled: true,
+        search_enabled: false,
+        tool_names: [],
+        deepseek_token: 'token-1',
+        pow_header: 'pow-1',
+        payload: { chat_session_id: 'session-1', prompt: 'hello', ref_file_ids: ['file-1'] },
+      });
+    }
+    if (textURL.includes('__stream_pow=1')) {
+      return jsonResponse({ pow_header: 'pow-retry' });
+    }
+    if (textURL.includes('__stream_switch=1')) {
+      return jsonResponse({
+        session_id: 'session-2',
+        lease_id: 'lease-test',
+        model: 'gpt-test',
+        final_prompt: 'hello',
+        thinking_enabled: true,
+        search_enabled: false,
+        tool_names: [],
+        deepseek_token: 'token-2',
+        pow_header: 'pow-2',
+        payload: { chat_session_id: 'session-2', prompt: 'hello', ref_file_ids: ['file-2'] },
+      });
+    }
+    if (textURL.includes('__stream_release=1')) {
+      return jsonResponse({ success: true });
+    }
+    if (textURL === 'https://chat.deepseek.com/api/v0/chat/completion') {
+      completionBodies.push(JSON.parse(String(init.body)));
+      completionAuth.push(init.headers.authorization);
+      completionCalls += 1;
+      if (completionCalls <= 2) {
+        return sseResponse([`data: {"response_message_id":${40 + completionCalls},"p":"response/thinking_content","v":"plan"}\n\n`, 'data: [DONE]\n\n']);
+      }
+      return sseResponse(['data: {"p":"response/content","v":"visible"}\n\n', 'data: [DONE]\n\n']);
+    }
+    throw new Error(`unexpected fetch url: ${textURL}`);
+  };
+  try {
+    const req = new MockStreamRequest();
+    const res = new MockStreamResponse();
+    const payload = { model: 'gpt-test', stream: true };
+    await handleVercelStream(req, res, Buffer.from(JSON.stringify(payload)), payload);
+    const frames = parseSSEDataFrames(res.bodyText());
+    const parsed = frames.filter((frame) => frame !== '[DONE]').map((frame) => JSON.parse(frame));
+    assert.equal(fetchURLs.filter((url) => url.includes('__stream_switch=1')).length, 1);
+    assert.equal(completionBodies.length, 3);
+    assert.match(completionBodies[1].prompt, /Previous reply had no visible output/);
+    assert.equal(completionBodies[1].parent_message_id, 41);
+    assert.equal(completionBodies[2].prompt, 'hello');
+    assert.deepEqual(completionBodies[2].ref_file_ids, ['file-2']);
+    assert.deepEqual(completionAuth, ['Bearer token-1', 'Bearer token-1', 'Bearer token-2']);
+    assert.equal(parsed.at(-1).choices[0].finish_reason, 'stop');
+  } finally {
+    global.fetch = originalFetch;
+  }
+});
+
 test('vercel stream coalesces many small content deltas while keeping one choice', async () => {
   const lines = Array.from({ length: 100 }, () => `data: ${JSON.stringify({ p: 'response/content', v: '字' })}\n\n`);
   lines.push('data: [DONE]\n\n');
@@ -653,6 +727,17 @@ test('parseChunkForContent strips leaked thought control markers from content',
   assert.deepEqual(parsed.parts, [{ text: 'ABC', type: 'text' }]);
 });
 
+test('parseChunkForContent strips fullwidth-delimited leaked control markers from content', () => {
+  const fw = '\uff5c';
+  const chunk = {
+    p: 'response/content',
+    v: `<${fw}begin▁of▁sentence${fw}>A<${fw}▁of▁thought${fw}>B<${fw} end_of_sentence ${fw}>C`,
+  };
+  const parsed = parseChunkForContent(chunk, false, 'text');
+  assert.equal(parsed.finished, false);
+  assert.deepEqual(parsed.parts, [{ text: 'ABC', type: 'text' }]);
+});
+
 test('parseChunkForContent detects content_filter status and ignores upstream output tokens', () => {
   const chunk = {
     p: 'response',