@raolin2025/claude-code-node 2.8.7 → 2.8.9

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@raolin2025/claude-code-node",
-  "version": "2.8.7",
+  "version": "2.8.9",
   "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/core/cli.js

@@ -732,9 +732,19 @@ const systemPrompt = cliArgs.systemPrompt || DEFAULT_SYSTEM_PROMPT
         case 'window': {
           const arg = rest[0] ? rest.join(' ').trim() : ''
           if (!arg) {
-            // 无参数：显示当前窗口
+            // 无参数：显示当前窗口 + 真实 context 用量
+            // 用实时估算（estimateMessages）反映"当前 context 填充了多少窗口"，
+            // 而非依赖 usage 上报的 inputTokens——本地网关不返回 usage 时
+            // inputTokens 恒为 0，无法反映真实填充情况。
+            const estTokens = engine.tokenBudget.estimateMessages(engine.state.messages)
+            const windowTokens = engine.tokenBudget.maxTokens
+            const reserved = engine.tokenBudget.reservedForOutput || 0
+            const pct = windowTokens > 0
+              ? Math.min(100, Math.round((estTokens / (windowTokens - reserved)) * 100))
+              : 0
             console.log(`Context window: ${formatTokens(tokenBudget.maxTokens)} (${windowSourceLabel(windowSource)})`)
-            console.log(`  Input used: ${tokenBudget.inputTokens.toLocaleString()} / ${tokenBudget.maxTokens.toLocaleString()} (${tokenBudget.usagePercent}%)`)
+            console.log(`  Context used: ${estTokens.toLocaleString()} / ${windowTokens.toLocaleString()} (${pct}% of usable window)`)
+            console.log(`  (real-time estimate; API-reported input: ${tokenBudget.inputTokens.toLocaleString()} tok)`)
             const manual = config.get('maxBudgetTokens') || 0
             if (manual > 0) console.log(`  Manual override: ${formatTokens(manual)} (persisted in config)`)
             break
@@ -893,6 +903,11 @@ const systemPrompt = cliArgs.systemPrompt || DEFAULT_SYSTEM_PROMPT
           break
         case 'budget': {
           console.log(tokenBudget.format())
+          // 追加实时估算的 context 用量（本地网关不返回 usage 时 inputTokens 为 0，无法反映真实填充）
+          const estTokens = engine.tokenBudget.estimateMessages(engine.state.messages)
+          const winTk = engine.tokenBudget.maxTokens
+          const pct = winTk > 0 ? Math.min(100, Math.round((estTokens / winTk) * 100)) : 0
+          console.log(`  Context (est): ${estTokens.toLocaleString()} / ${winTk.toLocaleString()} (${pct}% of window)`)
           console.log(`Context window: ${formatTokens(tokenBudget.maxTokens)} (${windowSourceLabel(windowSource)}) | trigger: ${Math.round((tokenBudget.maxTokens * 0.8)).toLocaleString()} (80%)`)
           break
         }

package/src/core/query-engine.js

@@ -356,6 +356,20 @@ export class QueryEngine {
       ...(useStream && { stream: true }),
     }
 
+    // [debug] 打印实际发送请求的关键信息，用于排查"模型为何不调用工具"
+    if (this.config.verbose) {
+      const sysCount = messages.filter(m => m.role === 'system').length
+      const toolCallsInHistory = messages.filter(m => m.role === 'assistant' && m.tool_calls).length
+      console.error(`[debug] → ${url}`)
+      console.error(`[debug]   model=${body.model} | messages=${messages.length} (system=${sysCount}, user=${messages.filter(m=>m.role==='user').length}, assistant=${messages.filter(m=>m.role==='assistant').length}, tool=${messages.filter(m=>m.role==='tool').length}) | tools=${tools.length} | max_tokens=${body.max_tokens}`)
+      console.error(`[debug]   history tool_calls=${toolCallsInHistory} | 首条=${messages[0]?.role}:${String(messages[0]?.content).slice(0,40)} | 末条=${messages[messages.length-1]?.role}:${String(messages[messages.length-1]?.content).slice(0,40)}`)
+      if (tools.length) {
+        console.error(`[debug]   工具名: ${tools.map(t=>t.function.name).join(', ')}`)
+      } else {
+        console.error(`[debug]   ⚠️ tools 为空！请求未携带工具定义，模型不会调用工具`)
+      }
+    }
+
     // DeepSeek V4: 默认启用 thinking mode → 要求回传 reasoning_content
     // 这里显式关闭，避免 tool call 场景下的 400 错误
     // thinking 参数是 DeepSeek 私有扩展，直接放在 body 顶层