Gate reasoning history replay

Author: jahooma

common/src/constants/freebuff-models.ts

@@ -49,12 +49,6 @@ export const FREEBUFF_MODELS = [
     tagline: 'Deepest, 1/day',
     availability: 'always',
   },
-  {
-    id: FREEBUFF_DEEPSEEK_V4_PRO_MODEL_ID,
-    displayName: 'DeepSeek V4 Pro',
-    tagline: 'Experimental',
-    availability: 'always',
-  },
   {
     id: FREEBUFF_MINIMAX_MODEL_ID,
     displayName: 'MiniMax M2.7',

packages/agent-runtime/src/constants.ts

@@ -9,3 +9,7 @@ export const globalStopSequence = `${JSON.stringify(endsAgentStepParam)}`
  * to diff sequential requests and find what's breaking prompt caching.
  */
 export const CACHE_DEBUG_FULL_LOGGING = false
+
+// Keep disabled by default to preserve mainline behavior until reasoning-token
+// replay has been tested more thoroughly.
+export const INCLUDE_REASONING_IN_MESSAGE_HISTORY = false

packages/agent-runtime/src/tools/stream-parser.ts

@@ -8,6 +8,7 @@ import {
 import { generateCompactId } from '@codebuff/common/util/string'
 
 import { processStreamWithTools } from '../tool-stream-parser'
+import { INCLUDE_REASONING_IN_MESSAGE_HISTORY } from '../constants'
 import {
   executeCustomToolCall,
   executeToolCall,
@@ -276,7 +277,7 @@ export async function processStream(
       }
 
       if (chunk.type === 'reasoning') {
-        if (chunk.text) {
+        if (INCLUDE_REASONING_IN_MESSAGE_HISTORY && chunk.text) {
           assistantMessages.push(
             assistantMessage({ type: 'reasoning', text: chunk.text }),
           )