@raolin2025/claude-code-node 2.8.16 → 2.8.17

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@raolin2025/claude-code-node",
-  "version": "2.8.16",
+  "version": "2.8.17",
   "description": "Node.js AI Code Agent CLI - Zero dependencies, pure JavaScript, security hardened, multi-channel notifications, Telegram & QQ Bot remote programming, rich media upload, multi-account management",
   "type": "module",
   "main": "src/core/index.js",

package/src/__tests__/small-model.test.js

@@ -226,3 +226,29 @@ test('max_tokens：绝不超过窗口一半（防超窗）', async () => {
   const qe2 = new QueryEngine({ tokenBudget: new TokenBudget({ maxTokens: 10000 }), outputRatio: 1.0 })
   assert.ok(qe2._computeMaxOutputTokens() <= 5000, `max_tokens=${qe2._computeMaxOutputTokens()} 不应超过窗口一半 5000`)
 })
+
+// ---- _buildRequest 把 system 统一前置（防 llama.cpp 500）----
+test('_buildRequest：把中间 system 统一前置到开头', async () => {
+  const { QueryEngine } = await import('../core/query-engine.js')
+  const qe = new QueryEngine({ systemPrompt: 'SYS-ROOT', tools: [] })
+  // 模拟 state.messages 里 system 摘要被插到中间（折叠/恢复导致）
+  const messages = [
+    { role: 'user', content: 'u1' },
+    { role: 'assistant', content: 'a1' },
+    { role: 'system', content: '[Context Summary] 中间摘要' },
+    { role: 'user', content: 'u2' },
+  ]
+  const req = qe._buildRequest(messages)
+  // 验证：所有 system 在开头，中间无 system
+  let seenBody = false, badSystem = false
+  for (const m of req) {
+    if (m.role !== 'system') seenBody = true
+    else if (seenBody) badSystem = true
+  }
+  assert.equal(badSystem, false, '中间不应出现 system（否则 llama.cpp 报 500）')
+  // 第一条是 system
+  assert.equal(req[0].role, 'system')
+  // 原始顺序（非 system 部分）应保持
+  const nonSys = req.filter(m => m.role !== 'system').map(m => m.role)
+  assert.deepEqual(nonSys, ['user', 'assistant', 'user'])
+})

package/src/core/query-engine.js

@@ -358,10 +358,25 @@ export class QueryEngine {
     }
 
     // 历史消息 — 转换为 OpenAI 兼容格式
+    // 关键：把所有 system 消息【收集到一起放到最前面】，而不是原样透传。
+    // 原因：llama.cpp/Jinja 严格要求 system 消息必须在对话开头，中间出现 system
+    // 会报 "System message must be at the beginning" (500)。
+    // 折叠/压缩/引导注入可能在 state.messages 中间留下 system 摘要消息，
+    // 这里统一前置，彻底规避该错误。
+    const systemMsgs = []
+    const bodyMsgs = []
     for (const msg of messages) {
       if (msg.role === 'system') {
-        request.push({ role: 'system', content: msg.content })
-      } else if (msg.role === 'user') {
+        systemMsgs.push(msg)
+      } else {
+        bodyMsgs.push(msg)
+      }
+    }
+    for (const msg of systemMsgs) {
+      request.push({ role: 'system', content: msg.content })
+    }
+    for (const msg of bodyMsgs) {
+      if (msg.role === 'user') {
         request.push({ role: 'user', content: this._buildUserContent(msg) })
       } else if (msg.role === 'assistant') {
         // 构建 assistant 消息基础