opencode-thinking-fix 1.1.3 → 1.1.5

package/README.md

@@ -10,6 +10,8 @@ npm install opencode-thinking-fix
 ```
 
 > Fix for the `reasoning_content` 400 error that kills multi-turn conversations with DeepSeek, Kimi, GLM, MiMo, and MiniMax-M3 in OpenCode.
+>
+> Docs: [OpenCode Plugins](https://opencode.ai/docs/plugins)
 
 ---
 
@@ -17,13 +19,15 @@ npm install opencode-thinking-fix
 
 **This is an OpenCode plugin. Install it inside OpenCode — no terminal needed.**
 
-### Method 1: TUI (press Ctrl+P while OpenCode is running)
+### Method 1: TUI (press `Ctrl+P` while OpenCode is running)
 
-1. Press `Ctrl+P` to open the command palette
-2. Type `install plugin` and press Enter
-3. Press `Tab` to switch scope to **Global** (recommended — works across all projects)
-4. Type `opencode-thinking-fix`
-5. Press Enter — done. Restart OpenCode.
+1. Press `Ctrl+P` to open the command palette.
+2. Type `install plugin` and press `Enter`.
+3. Press `Tab` to switch the install scope to **Global** (recommended — works across all projects).
+4. Type `opencode-thinking-fix`.
+5. Press `Enter`. Restart OpenCode.
+
+You should see `[ThinkingFix] Plugin loaded — universal reasoning_content fix active` at startup.
 
 ### Method 2: CLI (shell command)
 
@@ -31,13 +35,15 @@ npm install opencode-thinking-fix
 opencode plugin opencode-thinking-fix
 ```
 
-For specific version:
+For a specific version:
 
 ```bash
-opencode plugin opencode-thinking-fix@1.1.0
+opencode plugin opencode-thinking-fix@1.1.4
 ```
 
-### Method 3: Manual config (add to opencode.json)
+Restart OpenCode after installing.
+
+### Method 3: Manual config (add to `opencode.json`)
 
 ```json
 {
@@ -51,7 +57,9 @@ Config file location:
 
 Restart OpenCode after adding. You should see `[ThinkingFix] Plugin loaded` at startup.
 
-## Table of Contents
+See also: [OpenCode plugin docs](https://opencode.ai/docs/plugins)
+
+---
 
 - [What problem this fixes](#what-problem-this-fixes)
 - [Option 1: Plugin (stops the crashes)](#option-1-plugin-stops-the-crashes)

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "opencode-thinking-fix",
-  "version": "1.1.3",
+  "version": "1.1.5",
   "description": "Fix DeepSeek/Kimi/GLM/MiMo reasoning_content 400 errors in OpenCode. Plugin + proxy + watchdog.",
   "license": "MIT",
   "author": "tbosancheros39",

package/proxy/proxy.js

@@ -155,9 +155,13 @@ function createStreamParser(sessionId, onComplete) {
       buffer += chunk.toString()
       const lines = buffer.split('\n')
       buffer = lines.pop() // keep incomplete last line
+      const outLines = []
 
       for (const line of lines) {
-        if (!line.startsWith('data: ')) continue
+        if (!line.startsWith('data: ')) {
+          outLines.push(line)
+          continue
+        }
         const data = line.slice(6).trim()
         if (data === '[DONE]') {
           // Flush accumulated reasoning for this turn
@@ -165,12 +169,16 @@ function createStreamParser(sessionId, onComplete) {
             onComplete(assistantIndex, reasoningBuffer)
             reasoningBuffer = ''
           }
+          outLines.push(line)
           continue
         }
         try {
           const parsed = JSON.parse(data)
           const delta = parsed.choices?.[0]?.delta
-          if (!delta) continue
+          if (!delta) {
+            outLines.push(line)
+            continue
+          }
 
           // reasoning_content delta (DeepSeek/Kimi/GLM/MiniMax)
           if (typeof delta.reasoning_content === 'string') {
@@ -194,6 +202,17 @@ function createStreamParser(sessionId, onComplete) {
             inReasoning = true
           }
 
+          // content may contain embedded <think> tags when reasoning_split is not
+          // honored (MiniMax fallback). Extract them into reasoningBuffer.
+          if (typeof delta.content === 'string') {
+            const thinkMatch = delta.content.match(/<think>([\s\S]*?)<\/think>/)
+            if (thinkMatch) {
+              reasoningBuffer += thinkMatch[1]
+              inReasoning = true
+              delta.content = delta.content.replace(/<think>[\s\S]*?<\/think>/, '').trimStart()
+            }
+          }
+
           // 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)
@@ -205,10 +224,16 @@ function createStreamParser(sessionId, onComplete) {
             }
             assistantIndex++
           }
+
+          // Re-serialize the possibly-modified chunk and forward it
+          outLines.push('data: ' + JSON.stringify(parsed))
         } catch {
-          // malformed SSE chunk, skip
+          // malformed SSE chunk, forward unchanged
+          outLines.push(line)
         }
       }
+
+      return outLines.join('\n') + '\n'
     },
     flush() {
       if (reasoningBuffer) {
@@ -240,16 +265,22 @@ const server = http.createServer((req, res) => {
     // Determine upstream route
     let modelName = ''
     let routeTarget = DEFAULT_ROUTE
+
+    // Always parse model name from body so provider-specific patches work
+    // even in fixed-upstream mode (OpenCode Go)
+    if (isPost && isJson && rawBody.length > 0) {
+      try {
+        const bodyJson = JSON.parse(rawBody.toString('utf8'))
+        modelName = bodyJson.model || ''
+      } catch { /* body not JSON yet */ }
+    }
+
     if (UPSTREAM_URL) {
       // Fixed upstream mode (e.g. OpenCode Go)
       routeTarget = { base: UPSTREAM_URL, reasoning: true }
-    } else if (isPost && isJson && rawBody.length > 0) {
+    } else {
       // Model-based routing mode
-      try {
-        const bodyJson = JSON.parse(rawBody.toString('utf8'))
-        modelName = bodyJson.model || ''
-        routeTarget = route(modelName)
-      } catch { /* body not JSON yet, use defaults */ }
+      routeTarget = route(modelName)
     }
     const upstream = url.parse(routeTarget.base)
     const shouldPatch = isPost && isJson && routeTarget.reasoning
@@ -262,10 +293,21 @@ const server = http.createServer((req, res) => {
         const bj = JSON.parse(bodyStr)
 
         // Strip unsupported params for Kimi/Moonshot thinking mode
+        // Moonshot hardcodes sampling params for thinking models and rejects
+        // explicit values. Docs: https://platform.kimi.ai/docs/api/models-overview
         const isKimi = modelName.toLowerCase().startsWith('kimi') || modelName.toLowerCase().startsWith('moonshot')
         if (isKimi) {
+          delete bj.temperature
           delete bj.top_p
           delete bj.top_k
+          delete bj.presence_penalty
+          delete bj.frequency_penalty
+          delete bj.n
+          // K2.7 thinking is always on; do not send thinking/reasoning_effort controls
+          if (modelName.toLowerCase().startsWith('kimi-k2.7')) {
+            delete bj.thinking
+            delete bj.reasoning_effort
+          }
         }
 
         // Inject reasoning_split for MiniMax models (avoids <think> tags in content)
@@ -318,8 +360,9 @@ const server = http.createServer((req, res) => {
         })
 
         proxyRes.on('data', chunk => {
-          parser.feed(chunk)
-          res.write(chunk) // forward chunk immediately (no buffering)
+          // parser.feed returns the (possibly modified) SSE text to forward
+          const forwardChunk = parser.feed(chunk)
+          res.write(forwardChunk)
         })
 
         proxyRes.on('end', () => {