fix(serve): skip thinking-only message fragments to prevent duplicate response in cs-cloud

normalizeMessages() splits a multi-block AssistantMessage (e.g. [thinking,
text]) into one message per content block. When serve mode emits both the
thinking-only fragment and the text message as separate session.message events,
cs-cloud's adaptMessageEvent creates two assistant messages (one via the
streaming path, one via the non-streaming path), causing the response to appear
twice in the UI.

Fix: skip thinking-only normalized messages in normalizeMessage(), since the
thinking content is already rendered via the stream_event path.

Co-Authored-By: claude-sonnet-4-6 <noreply@anthropic.com>
This commit is contained in:
Askhz 2026-05-15 17:58:53 +08:00
parent 2857e77065
commit 02899ce69f

View File

@ -26,7 +26,7 @@ import {
createFileStateCacheWithSizeLimit,
type FileStateCache,
} from './fileStateCache.js'
import { isNotEmptyMessage, normalizeMessages } from './messages.js'
import { isNotEmptyMessage, isThinkingMessage, normalizeMessages } from './messages.js'
import { expandPath } from './path.js'
import type {
inputSchema as permissionToolInputSchema,
@ -117,6 +117,13 @@ export function* normalizeMessage(message: Message): Generator<SDKMessage> {
if (!isNotEmptyMessage(_)) {
continue
}
// Skip thinking-only messages: normalizeMessages splits a multi-block
// AssistantMessage (e.g. [thinking, text]) into one message per block.
// Emitting the thinking-only fragment causes serve mode to send a second
// session.message event, which cs-cloud renders as a duplicate response.
if (isThinkingMessage(_)) {
continue
}
yield {
type: 'assistant',
message: _.message,