opencode-thinking-fix 1.1.1 → 1.1.2

package/README.md

@@ -163,6 +163,8 @@ Then point OpenCode at it — in your `opencode.json`:
 
 Zero npm dependencies. It uses `http`, `https`, and `url` — nothing else.
 
+**Interleaved thinking support:** GLM-5+ and MiniMax-M3 emit reasoning AFTER content in the same turn (interleaved thinking between tool calls). The proxy accumulates ALL reasoning across an entire assistant turn and flushes only on `finish_reason` — never on `delta.content` arrival. This prevents split/lost reasoning blocks.
+
 **Kimi K2.7 Code and OpenCode Go need this.** The rest of the models benefit from it but do not technically require it.
 
 ---
@@ -202,7 +204,7 @@ The plugin is the safety net. If the proxy goes down, the plugin still injects e
 | **Kimi K2.7 Code** | **Not enough alone** | **Required** | Needs real text |
 | GLM-5.x / Zhipu | Yes | Nice to have | Accepts `""` |
 | MiMo V2.5 / MiniMax | Yes | Nice to have | Accepts `""` (default mode embeds `<think>` in `content`) |
-| **MiniMax-M3** | Yes | **Recommended** | `reasoning_details[]` array format; ~40% quality loss if stripped |
+| **MiniMax-M3** | Yes | **Recommended** | `reasoning_details[]` array; ~40% quality loss if stripped. Proxy injects `reasoning_split:true` to keep thinking separate from content. |
 | OpenCode Go | Yes | Required | Uses `reasoning` field |
 | Qwen, GPT, Claude, Gemini, Llama, Mistral | No | No | No reasoning_content |
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "opencode-thinking-fix",
-  "version": "1.1.1",
+  "version": "1.1.2",
   "description": "Fix DeepSeek/Kimi/GLM/MiMo reasoning_content 400 errors in OpenCode. Plugin + proxy + watchdog.",
   "license": "MIT",
   "author": "tbosancheros39",

package/proxy/proxy.js

@@ -172,7 +172,7 @@ function createStreamParser(sessionId, onComplete) {
           const delta = parsed.choices?.[0]?.delta
           if (!delta) continue
 
-          // reasoning_content delta (DeepSeek/Kimi/GLM native)
+          // reasoning_content delta (DeepSeek/Kimi/GLM/MiniMax)
           if (typeof delta.reasoning_content === 'string') {
             reasoningBuffer += delta.reasoning_content
             inReasoning = true
@@ -194,16 +194,9 @@ function createStreamParser(sessionId, onComplete) {
             inReasoning = true
           }
 
-          // When content starts after reasoning, the reasoning block is done
-          if (typeof delta.content === 'string' && delta.content && inReasoning) {
-            if (reasoningBuffer) {
-              onComplete(assistantIndex, reasoningBuffer)
-              reasoningBuffer = ''
-              inReasoning = false
-            }
-          }
-
           // finish_reason means the assistant turn is complete
+          // Only flush here — NOT on content appearance (interleaved thinking
+          // emits reasoning AFTER content on GLM-5+ and MiniMax-M3)
           if (parsed.choices?.[0]?.finish_reason) {
             if (reasoningBuffer) {
               onComplete(assistantIndex, reasoningBuffer)
@@ -264,7 +257,17 @@ const server = http.createServer((req, res) => {
     // Patch request body if this is a reasoning provider
     let bodyToSend = rawBody
     if (shouldPatch && rawBody.length > 0) {
-      const patched = patchRequestBody(rawBody.toString('utf8'), sessionId)
+      // Inject reasoning_split for MiniMax models (avoids <think> tags in content)
+      const isMiniMax = modelName.toLowerCase().startsWith('minimax') || modelName.toLowerCase().startsWith('mimo')
+      let bodyStr = rawBody.toString('utf8')
+      if (isMiniMax) {
+        try {
+          const bj = JSON.parse(bodyStr)
+          bj.reasoning_split = true
+          bodyStr = JSON.stringify(bj)
+        } catch { /* pass */ }
+      }
+      const patched = patchRequestBody(bodyStr, sessionId)
       bodyToSend = Buffer.from(patched, 'utf8')
     }