From 80581deb3766b5372582e2670595974f1441f0fb Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E7=B4=AB=E5=BD=B1233?= <thzxx@users.noreply.gitee.com>
Date: Thu, 16 Jul 2026 16:44:04 +0800
Subject: [PATCH] =?UTF-8?q?v0.16.10:=20=E6=9C=80=E5=B0=8F=E5=BF=85?=
 =?UTF-8?q?=E8=A6=81=E6=80=A7=E6=B8=85=E7=90=86=20=E2=80=94=20=E5=88=A0?=
 =?UTF-8?q?=E9=99=A4=E9=9D=9E=E5=BF=85=E8=A6=81=E6=A0=A1=E9=AA=8C=E5=92=8C?=
 =?UTF-8?q?=E6=B3=A8=E5=85=A5=E6=8F=90=E7=A4=BA?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

按"让 AI 自己判断"原则，删除所有"工程师判断"性质的校验代码和注入提示，
只保留安全防护（路径沙箱/命令安全/参数消毒）和必要的上下文管理（压缩/截断）。

删除项:
- completion-gate.ts 整个文件（notThinking/toolResultReview/contextEfficiency/planModeCompletion）
- verifyToolResult 工具结果核验函数 + TOOLS_NEED_VERIFY 常量
- 跨轮次死循环检测器（recordLoopSignature/detectLoopDeadlock/resetLoopDeadlockDetector）
- R77 checkRateLimit 速率限制 + R87 isToolCircuitBroken 熔断器 + R104 isDuplicateToolResult 去重
- R56 目标对齐验证 + R63 速率限制 + R87 熔断器 + R104 去重检测 + R119 优先级排序
- agent-metrics recordCompletionGate + completionGatePassed + avgCompletionScore 相关代码
- context-manager 低价值关键词黑名单 + 快速摘要改用 user role
- LoopContext 的 completionGateFailCount/verifyWarnings 字段

修复项:
- R76 路径沙箱 replace bug：用 startsWith 前缀锚定替代 replace，避免子串误判
- R28 命令注入检测：缩小匹配范围，仅拦截命令替换中包含危险命令的情况

总计 13 文件变更，+46/-1124 行
---
 README.md                                |   6 +-
 package-lock.json                        |   4 +-
 package.json                             |   2 +-
 src/main/menu.ts                         |   2 +-
 src/renderer/index.html                  |   2 +-
 src/renderer/services/agent-engine.ts    | 311 +--------------
 src/renderer/services/agent-metrics.ts   |  17 -
 src/renderer/services/agent-safety.ts    | 458 +----------------------
 src/renderer/services/completion-gate.ts | 298 ---------------
 src/renderer/services/context-manager.ts |  14 +-
 src/renderer/services/sub-agent.ts       |  34 +-
 src/renderer/services/tool-registry.ts   |  10 +-
 src/renderer/types.d.ts                  |  12 -
 13 files changed, 46 insertions(+), 1124 deletions(-)
 delete mode 100644 src/renderer/services/completion-gate.ts

diff --git a/README.md b/README.md
index ef61a5a..6983e27 100644
--- a/README.md
+++ b/README.md
@@ -14,7 +14,7 @@
 </p>
 
 <p align="center">
-  <img src="https://img.shields.io/badge/version-v0.16.9-E8734A?style=flat-square" alt="version">
+  <img src="https://img.shields.io/badge/version-v0.16.10-E8734A?style=flat-square" alt="version">
   <img src="https://img.shields.io/badge/electron-33+-47848F?style=flat-square&logo=electron" alt="electron">
   <img src="https://img.shields.io/badge/typescript-5.7+-3178C6?style=flat-square&logo=typescript" alt="typescript">
   <img src="https://img.shields.io/badge/license-MIT-green?style=flat-square" alt="license">
@@ -253,7 +253,7 @@ npm start
 ELECTRON_MIRROR=https://npmmirror.com/mirrors/electron/ npm run dist
 ```
 
-产出：`release/Metona Ollama Setup v0.16.9.exe`
+产出：`release/Metona Ollama Setup v0.16.10.exe`
 
 ## 🛠️ 常用命令
 
@@ -501,7 +501,7 @@ npm start
 ELECTRON_MIRROR=https://npmmirror.com/mirrors/electron/ npm run dist
 ```
 
-Output: `release/Metona Ollama Setup v0.16.9.exe`
+Output: `release/Metona Ollama Setup v0.16.10.exe`
 
 ## 🛠️ Common Commands
 
diff --git a/package-lock.json b/package-lock.json
index 55aaeae..5fc3d96 100644
--- a/package-lock.json
+++ b/package-lock.json
@@ -1,12 +1,12 @@
 {
   "name": "metona-ollama-desktop",
-  "version": "0.16.9",
+  "version": "0.16.10",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "metona-ollama-desktop",
-      "version": "0.16.9",
+      "version": "0.16.10",
       "license": "MIT",
       "dependencies": {
         "ffmpeg-static": "^5.2.0",
diff --git a/package.json b/package.json
index 3cf76c6..a16cbd9 100644
--- a/package.json
+++ b/package.json
@@ -1,6 +1,6 @@
 {
   "name": "metona-ollama-desktop",
-  "version": "0.16.9",
+  "version": "0.16.10",
   "description": "Metona Ollama - TypeScript + Electron 桌面 AI 聊天客户端",
   "main": "dist/main/main.js",
   "author": "thzxx",
diff --git a/src/main/menu.ts b/src/main/menu.ts
index 0fb03c6..7235831 100644
--- a/src/main/menu.ts
+++ b/src/main/menu.ts
@@ -101,7 +101,7 @@ export function createMenu(): void {
             dialog.showMessageBox(mainWindow!, {
               type: 'info',
               title: '关于 Metona Ollama',
-              message: 'Metona Ollama Desktop v0.16.9',
+              message: 'Metona Ollama Desktop v0.16.10',
               detail: 'TypeScript + Electron Ollama AI 聊天客户端\n\nhttps://gitee.com/thzxx/metona-ollama',
               icon: getIconPath()
             });
diff --git a/src/renderer/index.html b/src/renderer/index.html
index b9100fa..77b9420 100644
--- a/src/renderer/index.html
+++ b/src/renderer/index.html
@@ -28,7 +28,7 @@
         <div class="header-left">
             <img class="logo" src="./assets/icons/llama.png" alt="logo" />
             <span class="app-title">Metona Ollama</span>
-            <span class="app-version">v0.16.9</span>
+            <span class="app-version">v0.16.10</span>
             <button class="icon-btn help-btn" id="btnHelp" title="使用帮助">
                 <svg viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2">
                     <circle cx="12" cy="12" r="10"/><path d="M9.09 9a3 3 0 0 1 5.83 1c0 2-3 3-3 3"/>
diff --git a/src/renderer/services/agent-engine.ts b/src/renderer/services/agent-engine.ts
index 87ba418..b0d7ba0 100644
--- a/src/renderer/services/agent-engine.ts
+++ b/src/renderer/services/agent-engine.ts
@@ -18,21 +18,14 @@ import {
 } from './tool-registry.js';
 import {
   compactOldToolResult,
-  setUserGoal,
   resetAllSafetyState,
-  checkRateLimit,
   classifyError,
   calculateBackoff,
   validatePathSandbox,
-  isToolCircuitBroken,
-  recordToolFailure,
-  recordToolSuccess,
   // R88: 工具结果元数据
   addResultMetadata,
   // R97: 错误模式学习
   recordErrorPattern,
-  // R104: 工具结果去重
-  isDuplicateToolResult,
   // R109: 工具参数消毒
   sanitizeToolArgs,
   // R113: 命令安全检查
@@ -60,9 +53,8 @@ import {
   // R100: Token 使用统计报告
   generateTokenReport, formatTokenReport,
 } from './context-manager.js';
-import { runCompletionGate } from './completion-gate.js';
 import { executeHooks } from './hooks.js';
-import { recordIteration, recordToolCall, recordCompletionGate, startSessionMetrics, endSessionMetrics } from './agent-metrics.js';
+import { recordIteration, recordToolCall, startSessionMetrics, endSessionMetrics } from './agent-metrics.js';
 import { getEffectiveNumCtx } from '../components/model-bar.js';
 import type {
   OllamaMessage,
@@ -185,58 +177,12 @@ const SIDE_EFFECT_TOOLS = new Set([
 ]);
 
 // ═══════════════════════════════════════════════════════════════
-// 跨轮次死循环检测
-// 追踪每轮工具调用签名，检测连续重复以防止模型陷入死循环
-// 与 R104（仅记录日志）不同，本机制在达到阈值时主动干预：
-//   - 连续 3 轮相同 → 注入事实性提示（不修改模型输出）
-//   - 注入提示后仍连续 5 轮相同 → 硬性熔断，强制终止
+// 跨轮次死循环检测已删除 — 硬熔断过于激进，软提示干扰 AI 判断
+// AI 应自行判断是否需要继续调用相同工具
 // ═══════════════════════════════════════════════════════════════
-const _loopToolSignatures: string[] = [];
-const LOOP_HISTORY_MAX = 10;
-let _deadlockHintInjected = false;
 // P1-E6 修复：待清理的记忆提取 timer 列表，finally 块统一清理
 const _pendingMemoryTimers: ReturnType<typeof setTimeout>[] = [];
 
-/** 记录本轮工具调用签名（每轮 handleThinking 成功后调用一次） */
-function recordLoopSignature(toolCalls: ToolCall[]): void {
-  if (toolCalls.length === 0) {
-    _loopToolSignatures.push('__no_tools__');
-    _deadlockHintInjected = false; // 无工具调用，模型已改变行为，重置提示标志
-    return;
-  }
-  const sig = toolCalls.map(tc =>
-    `${tc.function.name}:${JSON.stringify(tc.function.arguments, Object.keys(tc.function.arguments || {}).sort())}`
-  ).join('|');
-  // 签名变化时重置提示标志（模型已改变行为）
-  if (_loopToolSignatures.length > 0 && _loopToolSignatures[_loopToolSignatures.length - 1] !== sig) {
-    _deadlockHintInjected = false;
-  }
-  _loopToolSignatures.push(sig);
-  if (_loopToolSignatures.length > LOOP_HISTORY_MAX) {
-    _loopToolSignatures.shift();
-  }
-}
-
-/** 检测连续 N 轮相同的工具调用签名 */
-function detectLoopDeadlock(minCount: number): { detected: boolean; toolName: string; count: number } {
-  if (_loopToolSignatures.length < minCount) return { detected: false, toolName: '', count: 0 };
-  const recent = _loopToolSignatures.slice(-minCount);
-  if (recent[0] === '__no_tools__') return { detected: false, toolName: '', count: 0 };
-  const allSame = recent.every(s => s === recent[0]);
-  if (allSame) {
-    const firstSig = recent[0];
-    const toolName = firstSig.split(':')[0] || 'unknown';
-    return { detected: true, toolName, count: minCount };
-  }
-  return { detected: false, toolName: '', count: 0 };
-}
-
-/** 重置死循环检测状态（新会话开始时调用） */
-function resetLoopDeadlockDetector(): void {
-  _loopToolSignatures.length = 0;
-  _deadlockHintInjected = false;
-}
-
 /**
  * R3: 工具执行超时配置（毫秒）
  * 防止单个工具调用挂起导致整个 Agent Loop 卡死
@@ -1214,8 +1160,6 @@ function snapshotLoopContext(ctx: LoopContext): void {
     // P0-P3 修复：补全关键字段，至少保证运行时调试可见
     mode: ctx.mode,
     emergencyCompressCount: ctx.emergencyCompressCount,
-    completionGateFailCount: ctx.completionGateFailCount,
-    verifyWarnings: ctx.verifyWarnings,
     compressedThisCycle: ctx.compressedThisCycle,
     messageCount: ctx.messages.length,
   });
@@ -1306,11 +1250,9 @@ async function handleInit(
 ): Promise<void> {
   // 新一轮对话，清空工具缓存
   toolResultCache.clear();
-resetAllSafetyState(); // R51-R56: 重置所有安全状态
-resetLoopDeadlockDetector(); // 重置跨轮次死循环检测
-resetTokenBudget(); // R93: 重置 Token 预算追踪
-setTokenBudgetNumCtx(getEffectiveNumCtx()); // R93: 设置当前预算 numCtx
-setUserGoal(userContent); // R56: 设置用户原始目标
+  resetAllSafetyState(); // R51-R56: 重置所有安全状态
+  resetTokenBudget(); // R93: 重置 Token 预算追踪
+  setTokenBudgetNumCtx(getEffectiveNumCtx()); // R93: 设置当前预算 numCtx
 
   const modelSupportsTools = state.get<boolean>('modelSupportsTools', false);
   // R55: 语义工具检索 — 根据用户查询过滤相关工具，减少 token 占用
@@ -1520,134 +1462,8 @@ function extractPlanSteps(content: string): string[] {
   return steps.slice(0, 8);
 }
 
-/**
- * 通用工具结果核验 — 对所有写类工具执行后验证
- * 读取/搜索类工具已有返回结果作为验证，此处只核验会产生副作用的操作
- */
-const TOOLS_NEED_VERIFY = new Set([
-  'write_file', 'edit_file', 'delete_file', 'create_directory',
-  'move_file', 'copy_file', 'download_file',
-]);
-
-async function verifyToolResult(
-  toolName: string,
-  args: Record<string, unknown>,
-  result: Record<string, unknown>,
-): Promise<string | null> {
-  // P3 #13 修复：返回警告字符串供 Completion Gate 使用
-  if (!TOOLS_NEED_VERIFY.has(toolName) || !result.success) return null;
-
-  const bridge = window.metonaDesktop;
-  if (!bridge?.isDesktop) return null;
-
-  try {
-    switch (toolName) {
-      case 'write_file': {
-        const path = String(args.path || '');
-        const expected = String(args.content || '');
-        const read = await bridge.tool.execute('read_file', { path, mode: 'text' });
-        if (read.success) {
-          const actual = String(read.content || '');
-          if (actual.length === 0) {
-            logWarn(`核验 write_file: ${path} 内容为空`);
-            return `write_file 核验: ${path} 写入后内容为空`;
-          } else if (Math.abs(actual.length - expected.length) > expected.length * 0.5) {
-            logWarn(`核验 write_file: ${path} 期望 ${expected.length}B 实际 ${actual.length}B`);
-            return `write_file 核验: ${path} 写入内容大小不符 (期望 ${expected.length}B, 实际 ${actual.length}B)`;
-          } else {
-            logInfo(`核验 write_file ✅: ${path} (${actual.length}B)`);
-          }
-        }
-        break;
-      }
-      case 'edit_file': {
-        const path = String(args.path || '');
-        const newText = String(args.new_text || '');
-        const read = await bridge.tool.execute('read_file', { path, mode: 'text' });
-        if (read.success) {
-          const actual = String(read.content || '');
-          if (!actual.includes(newText)) {
-            logWarn(`核验 edit_file: ${path} 不包含期望的新内容`);
-            return `edit_file 核验: ${path} 编辑后文件不包含期望的新内容`;
-          } else {
-            logInfo(`核验 edit_file ✅: ${path}`);
-          }
-        }
-        break;
-      }
-      case 'delete_file': {
-        // 支持批量删除核验：检查 paths 数组和单个 path
-        const pathsToCheck: string[] = [];
-        if (args.paths && Array.isArray(args.paths)) {
-          pathsToCheck.push(...args.paths.map((p: unknown) => String(p)));
-        }
-        if (args.path) pathsToCheck.push(String(args.path));
-        for (const p of pathsToCheck) {
-          const info = await bridge.tool.execute('get_file_info', { path: p });
-          if (info.success) {
-            logWarn(`核验 delete_file: ${p} 仍然存在，删除未生效`);
-            return `delete_file 核验: ${p} 删除后仍然存在`;
-          } else {
-            logInfo(`核验 delete_file ✅: ${p} 已不存在`);
-          }
-        }
-        break;
-      }
-      case 'create_directory': {
-        const path = String(args.path || '');
-        const info = await bridge.tool.execute('get_file_info', { path });
-        if (info.success && (info as any).type === 'directory') {
-          logInfo(`核验 create_directory ✅: ${path}`);
-        } else {
-          logWarn(`核验 create_directory: ${path} 未成功创建`);
-          return `create_directory 核验: ${path} 目录未成功创建`;
-        }
-        break;
-      }
-      case 'move_file': {
-        const src = String(args.source || '');
-        const dest = String(args.destination || '');
-        const [srcInfo, destInfo] = await Promise.all([
-          bridge.tool.execute('get_file_info', { path: src }),
-          bridge.tool.execute('get_file_info', { path: dest }),
-        ]);
-        if (srcInfo.success) {
-          logWarn(`核验 move_file: 源文件 ${src} 仍然存在`);
-          return `move_file 核验: 源文件 ${src} 移动后仍然存在`;
-        } else if (destInfo.success) {
-          logInfo(`核验 move_file ✅: ${src} → ${dest}`);
-        } else {
-          logWarn(`核验 move_file: ${src}→${dest} 源和目标均不存在`);
-          return `move_file 核验: ${src}→${dest} 移动后源和目标均不存在`;
-        }
-        break;
-      }
-      case 'copy_file': {
-        const dest = String(args.destination || '');
-        const destInfo = await bridge.tool.execute('get_file_info', { path: dest });
-        if (destInfo.success) {
-          logInfo(`核验 copy_file ✅: ${dest} (${(destInfo as any).size}B)`);
-        } else {
-          logWarn(`核验 copy_file: 目标 ${dest} 不存在`);
-          return `copy_file 核验: 目标 ${dest} 复制后不存在`;
-        }
-        break;
-      }
-      case 'download_file': {
-        const dest = String(args.destination || '');
-        const info = await bridge.tool.execute('get_file_info', { path: dest });
-        if (info.success && (info as any).size > 0) {
-          logInfo(`核验 download_file ✅: ${dest} (${(info as any).size}B)`);
-        } else {
-          logWarn(`核验 download_file: ${dest} 不存在或为空`);
-          return `download_file 核验: ${dest} 下载后不存在或为空`;
-        }
-        break;
-      }
-    }
-  } catch { /* 核验失败不阻塞主流程 */ }
-  return null;
-}
+// verifyToolResult 已删除 — 工具结果核验过于粗糙，误报边界情况
+// AI 应基于工具返回的 success/error 字段自行判断结果有效性
 
 /**
  * THINKING 状态：调用 LLM 流式
@@ -1668,33 +1484,6 @@ async function handleThinking(
   ctx.thinking = '';
   ctx.toolCalls.length = 0;
 
-  // ── 跨轮次死循环检测 ──
-  // 硬性熔断：已注入提示后仍连续 5 轮相同 → 强制终止
-  if (_deadlockHintInjected) {
-    const hardBreak = detectLoopDeadlock(5);
-    if (hardBreak.detected) {
-      logWarn(`死循环熔断: 连续 ${hardBreak.count} 轮相同工具调用 (${hardBreak.toolName})，已注入提示但仍未改善，强制终止`);
-      const fallbackReply = `已执行工具调用但模型未能生成最终回复。最后执行的工具: ${hardBreak.toolName}。请尝试重新描述任务或切换模型。`;
-      ctx.content = fallbackReply;
-      callbacks.onDone(fallbackReply, ctx.allToolRecords.length > 0 ? ctx.allToolRecords : undefined, makeStats(ctx));
-      transition(ctx, S.TERMINATED);
-      return;
-    }
-  }
-  // 软性提示：连续 3 轮相同工具调用 → 注入事实性提示（不修改模型输出）
-  if (!_deadlockHintInjected) {
-    const softWarn = detectLoopDeadlock(3);
-    if (softWarn.detected) {
-      logWarn(`死循环提示: 连续 ${softWarn.count} 轮相同工具调用 (${softWarn.toolName})，注入事实性提示`);
-      ctx.messages.push({
-        role: 'user',
-        content: `[系统提示] 你已连续 ${softWarn.count} 轮调用相同的工具 (${softWarn.toolName}) 并获得相同结果。任务可能已完成，请基于已有工具结果直接回复用户，不要再调用工具。`,
-        ephemeral: true,
-      });
-      _deadlockHintInjected = true;
-    }
-  }
-
   // ── Plan Mode 执行进度注入 — 仅在进度变化时注入，避免每轮重复 ──
   if (ctx.mode === 'plan' && ctx.loopCount >= 1) {
     const tracker = getPlanTracker();
@@ -1970,9 +1759,6 @@ async function handleThinking(
   ctx.loopPromptEvalCount = 0;
   ctx.loopInferenceNs = 0;
 
-  // 记录本轮工具调用签名（用于跨轮次死循环检测）
-  recordLoopSignature(ctx.toolCalls);
-
   transition(ctx, S.PARSING);
 }
 
@@ -2108,29 +1894,8 @@ async function handleExecuting(
       }, null];
     }
 
-    // R77: 工具调用速率限制 — 防止快速连续调用同一工具浪费资源
-    const rateLimitResult = checkRateLimit(call.function.name);
-    if (!rateLimitResult.allowed) {
-      const waitSec = Math.ceil(rateLimitResult.retryAfterMs / 1000);
-      logWarn(`R77: 速率限制拦截: ${call.function.name}，需等待 ${waitSec}s`);
-      return [{
-        name: call.function.name, arguments: call.function.arguments,
-        result: { success: false, error: `工具「${call.function.name}」调用频率过高，请等待 ${waitSec} 秒后重试。` },
-        status: 'error' as const, timestamp: Date.now()
-      }, null];
-    }
-
-    // R87: 工具熔断器检查 — 连续失败后自动禁用工具
-    const circuitBreaker = isToolCircuitBroken(call.function.name);
-    if (circuitBreaker.broken) {
-      const waitSec = Math.ceil(circuitBreaker.remainingMs / 1000);
-      logWarn(`R87: 熔断器拦截: ${call.function.name}，冷却中（还需 ${waitSec}s）`);
-      return [{
-        name: call.function.name, arguments: call.function.arguments,
-        result: { success: false, error: `工具「${call.function.name}」因连续失败已被暂时禁用，请等待 ${waitSec} 秒后重试，或使用其他方法。` },
-        status: 'error' as const, timestamp: Date.now()
-      }, null];
-    }
+    // R77/R87 已删除：速率限制 + 熔断器 — 剥夺 AI 试错空间，误伤密集型任务
+    // AI 应基于工具返回的错误信息自行调整策略
 
     // R79: 路径沙箱验证 — 确保文件操作不超出工作空间边界
     const FILE_PATH_TOOLS = new Set([
@@ -2221,12 +1986,6 @@ async function handleExecuting(
           state.get<AbortController | null>(KEYS.ABORT_CONTROLLER)?.signal)
           .catch(err => ({ success: false, error: err?.message || String(err) }) as ToolResult);
         logToolResult(call.function.name, result.success, result.success ? undefined : result.error);
-        // R87: 记录工具成功/失败到熔断器
-        if (result.success) {
-          recordToolSuccess(call.function.name);
-        } else {
-          recordToolFailure(call.function.name);
-        }
         // 记录 write_file 成功路径及内容指纹，供 FileWriteDedup Hook 做内容级去重
         if (call.function.name === 'write_file' && result.success && call.function.arguments?.path) {
           const { addWrittenFile } = await import('./hooks.js');
@@ -2310,13 +2069,6 @@ async function handleExecuting(
       ctx.allToolRecords.push(record);
       // R88: 为工具结果添加元数据提示（token 估算）
       const formattedResult = addResultMetadata(formatToolResultForModel(record.name, record.result!));
-      // R104: 检测重复工具结果（仅记录日志，不修改结果内容）
-      if (record.status === 'success') {
-        const dedup = isDuplicateToolResult(record.name, formattedResult);
-        if (dedup.duplicate) {
-          logInfo(`R104: 检测到重复工具结果: ${record.name}，已记录`);
-        }
-      }
       // R92: 工具结果格式标准化 — 添加统一头信息（工具名、状态、耗时、结果大小）
       const resultDuration = Date.now() - record.timestamp;
       const resultSize = formattedResult.length;
@@ -2333,15 +2085,6 @@ async function handleExecuting(
       if (cacheKey) setToolCache(cacheKey, { result: record.result!, timestamp: Date.now() });
       // 记录度量
       recordToolCall(record.name, record.status, Date.now() - record.timestamp);
-      // ── 通用工具结果核验 — 所有写类工具执行后验证 ──
-      // P3 #13 修复：收集核验警告到 ctx.verifyWarnings，供 Completion Gate 使用
-      if (record.status === 'success') {
-        const warning = await verifyToolResult(record.name, record.arguments, record.result!);
-        if (warning) {
-          ctx.verifyWarnings.push(warning);
-          logWarn(warning);
-        }
-      }
       // ── post_tool Hook ──
       executeHooks('post_tool', ctx, { toolName: record.name, toolArgs: record.arguments, toolResult: record.result! }).catch(err => logWarn('post_tool hook 异常', String(err)));
       // P0 修复: 使用参数匹配而非仅工具名匹配，避免同批次同名工具的错误关联
@@ -2695,35 +2438,9 @@ async function handleReflecting(
     }
   } catch { /* Hook 异常不影响主流程 */ }
 
-  try {
-    const gateResult = await runCompletionGate(ctx);
-    if (!gateResult.passed) {
-      recordCompletionGate(false);
-      logWarn(`Completion Gate: ${gateResult.reason}`);
-
-      // P1 #5 修复：Gate 失败时注入纠正提示并重试（最多 1 次），而非仅记录日志
-      const MAX_GATE_RETRIES = 1;
-      if (ctx.completionGateFailCount < MAX_GATE_RETRIES && ctx.loopCount < ctx.maxLoops) {
-        ctx.completionGateFailCount++;
-        logInfo(`Completion Gate: 注入纠正提示，重试 (${ctx.completionGateFailCount}/${MAX_GATE_RETRIES})`);
-        ctx.messages.push({
-          role: 'user',
-          content: `[系统提示] ${gateResult.reason}\n请基于已有信息补充完整你的回答，不要重复之前的工具调用。`,
-          ephemeral: true,
-        });
-        transition(ctx, S.THINKING);
-        return;
-      } else {
-        logWarn(`Completion Gate: 已达重试上限或循环上限，接受当前结果`);
-      }
-    } else {
-      recordCompletionGate(true);
-    }
-    // R90: 记录完成门控评分到日志
-    if (gateResult.score < 100) {
-      logInfo(`R90: Completion Gate 评分: ${gateResult.score}/100`);
-    }
-  } catch { /* Gate 异常不影响主流程 */ }
+  // Completion Gate 已删除 — 所有检查项（notThinking/toolResultReview/contextEfficiency/planModeCompletion）均已删除
+  // AI 应基于工具结果和上下文自行判断是否需要继续或完成
+  // Plan Mode 下 AI 可通过 formatPlanStatus() 看到剩余步骤，应自行决定是否继续
 
   logInfo('模型停止工具调用 → ReAct Agent Loop 结束');
   // ── 保存 Plan Mode 最终进度到状态，供下一轮 Loop 注入历史 ──
@@ -2881,8 +2598,6 @@ export async function runAgentLoop(
     startTime: Date.now(),
     lastLoopStats: undefined,
     emergencyCompressCount: 0,
-    completionGateFailCount: 0,
-    verifyWarnings: [],
     compressedThisCycle: false,
   };
 
diff --git a/src/renderer/services/agent-metrics.ts b/src/renderer/services/agent-metrics.ts
index 7d3644a..fc899cc 100644
--- a/src/renderer/services/agent-metrics.ts
+++ b/src/renderer/services/agent-metrics.ts
@@ -37,7 +37,6 @@ interface SessionMetrics {
   toolCalls: Array<{ name: string; status: string; duration?: number }>;
   totalInputTokens: number;
   totalOutputTokens: number;
-  completionGatePassed: boolean;
   errorPatterns: string[];
 }
 
@@ -55,7 +54,6 @@ export function startSessionMetrics(sessionId: string, model: string): void {
     toolCalls: [],
     totalInputTokens: 0,
     totalOutputTokens: 0,
-    completionGatePassed: false,
     errorPatterns: [],
   };
   logDebug(`Agent Metrics: 开始采集会话 ${sessionId}`);
@@ -82,12 +80,6 @@ export function recordToolCall(name: string, status: string, durationMs?: number
   }
 }
 
-/** 完成门控结果 */
-export function recordCompletionGate(passed: boolean): void {
-  if (!currentMetrics) return;
-  currentMetrics.completionGatePassed = passed;
-}
-
 /** 结束会话度量 */
 export function endSessionMetrics(): SessionMetrics | null {
   if (!currentMetrics) return null;
@@ -124,7 +116,6 @@ function persistMetricsHistory(): void {
       iterations: m.totalIterations,
       toolCallCount: m.toolCalls.length,
       toolSuccessRate: getToolSuccessRate(m),
-      completionGatePassed: m.completionGatePassed,
       errorPatterns: m.errorPatterns,
       inputTokens: m.totalInputTokens,
       outputTokens: m.totalOutputTokens,
@@ -152,7 +143,6 @@ export function loadMetricsHistory(): void {
         toolCalls: [],
         totalInputTokens: item.inputTokens || 0,
         totalOutputTokens: item.outputTokens || 0,
-        completionGatePassed: item.completionGatePassed || false,
         errorPatterns: item.errorPatterns || [],
       });
     }
@@ -192,7 +182,6 @@ export function aggregateMetrics(): AgentMetrics {
       totalSessions: 0,
       avgIterationsPerTask: 0,
       toolSuccessRate: 0,
-      avgCompletionScore: 0,
       frequentErrors: [],
       tokenEfficiency: 0,
       collectedAt: Date.now(),
@@ -221,7 +210,6 @@ export function aggregateMetrics(): AgentMetrics {
     totalSessions: history.length,
     avgIterationsPerTask: Math.round(totalIterations / history.length),
     toolSuccessRate: allToolCalls.length > 0 ? successCalls / allToolCalls.length : 0,
-    avgCompletionScore: history.filter(m => m.completionGatePassed).length / history.length,
     frequentErrors,
     tokenEfficiency: totalInputTokens > 0 ? totalOutputTokens / totalInputTokens : 0,
     collectedAt: Date.now(),
@@ -330,7 +318,6 @@ export function formatMetricsReport(metrics: AgentMetrics): string {
     `总会话数:         ${metrics.totalSessions}`,
     `平均迭代/任务:    ${metrics.avgIterationsPerTask}`,
     `工具成功率:       ${formatPercent(metrics.toolSuccessRate)}`,
-    `完成门控通过率:   ${formatPercent(metrics.avgCompletionScore)}`,
     `Token 效率:       ${formatPercent(metrics.tokenEfficiency)}`,
     `高频错误:         ${metrics.frequentErrors.length > 0 ? metrics.frequentErrors.map(e => `${e.pattern}(${e.count}次)`).join(', ') : '无'}`,
   ].join('\n');
@@ -347,7 +334,6 @@ export function exportMetricsJSON(): string {
       sessions_total: metrics.totalSessions,
       avg_iterations_per_task: metrics.avgIterationsPerTask,
       tool_success_rate: parseFloat(metrics.toolSuccessRate.toFixed(4)),
-      completion_gate_pass_rate: parseFloat(metrics.avgCompletionScore.toFixed(4)),
       token_efficiency: parseFloat(metrics.tokenEfficiency.toFixed(4)),
       top_errors: metrics.frequentErrors.slice(0, 5),
       improvement_suggestions: generateImprovementSuggestions().slice(0, 3).map(s => ({
@@ -373,9 +359,6 @@ export function exportMetricsPrometheus(): string {
   lines.push('# HELP metona_tool_success_rate Tool call success rate (0-1)');
   lines.push('# TYPE metona_tool_success_rate gauge');
   lines.push(`metona_tool_success_rate ${metrics.toolSuccessRate.toFixed(4)}`);
-  lines.push('# HELP metona_completion_gate_pass_rate Completion gate pass rate (0-1)');
-  lines.push('# TYPE metona_completion_gate_pass_rate gauge');
-  lines.push(`metona_completion_gate_pass_rate ${metrics.avgCompletionScore.toFixed(4)}`);
   lines.push('# HELP metona_token_efficiency Token output/input efficiency ratio');
   lines.push('# TYPE metona_token_efficiency gauge');
   lines.push(`metona_token_efficiency ${metrics.tokenEfficiency.toFixed(4)}`);
diff --git a/src/renderer/services/agent-safety.ts b/src/renderer/services/agent-safety.ts
index 3003b21..f8a6f3a 100644
--- a/src/renderer/services/agent-safety.ts
+++ b/src/renderer/services/agent-safety.ts
@@ -97,227 +97,10 @@ export function detectConsecutiveIdentical(minCount: number): { detected: boolea
 }
 
 // ═══════════════════════════════════════════════════════════════
-// R56: 目标对齐验证
+// R56 已删除：目标对齐验证 — 关键词匹配粗糙，反复注入干扰 AI 判断
+// R63 已删除：工具调用速率限制 — 剥夺 AI 试错空间，误伤密集型任务
 // ═══════════════════════════════════════════════════════════════
 
-/** 用户原始目标（在 handleInit 时设置） */
-let _userGoal: string = '';
-
-/** 设置用户原始目标 */
-export function setUserGoal(goal: string): void {
-  _userGoal = goal.slice(0, 500); // 限制长度防止占用过多内存
-}
-
-/** 获取用户原始目标 */
-export function getUserGoal(): string {
-  return _userGoal;
-}
-
-/** 重置用户目标 */
-export function resetUserGoal(): void {
-  _userGoal = '';
-}
-
-/** 目标对齐检测结果 */
-export interface GoalAlignmentResult {
-  aligned: boolean;
-  reason: string;
-  suggestion?: string;
-}
-
-/**
- * R56: 检测 Agent 是否偏离用户原始目标
- * 
- * 检测策略：
- * 1. 提取用户目标中的关键动词和名词
- * 2. 检查最近 5 轮的工具调用是否与目标关键词相关
- * 3. 如果连续多轮工具调用都与目标无关，判定为偏离
- * 
- * @param recentToolNames 最近 N 轮的工具调用名称列表
- * @param recentContent 最近 N 轮的模型输出内容摘要
- * @param loopCount 当前循环计数
- */
-export function checkGoalAlignment(
-  recentToolNames: string[],
-  recentContent: string,
-  loopCount: number,
-): GoalAlignmentResult {
-  if (!_userGoal || _userGoal.length < 10) {
-    return { aligned: true, reason: '目标太短，跳过对齐检测' };
-  }
-
-  // 仅在循环数 >= 6 时开始检测（前几轮可能是探索阶段）
-  if (loopCount < 6) {
-    return { aligned: true, reason: '前期探索阶段，跳过检测' };
-  }
-
-  const goalLower = _userGoal.toLowerCase();
-  
-  // 提取目标关键词
-  const goalKeywords = extractGoalKeywords(_userGoal);
-  
-  // 检查最近内容是否包含目标关键词
-  const contentLower = recentContent.toLowerCase();
-  const contentMatches = goalKeywords.filter(kw => contentLower.includes(kw.toLowerCase()));
-  
-  // 检查最近的工具调用是否与目标相关
-  // 每种工具与特定目标类型的关联度
-  const toolGoalRelevance = assessToolGoalRelevance(recentToolNames, goalLower);
-  
-  // 如果最近内容中包含目标关键词，认为是对齐的
-  if (contentMatches.length >= 1) {
-    return { aligned: true, reason: `内容中包含目标关键词: ${contentMatches.join(', ')}` };
-  }
-  
-  // 如果工具调用与目标高度相关，认为是对齐的
-  if (toolGoalRelevance.relevant) {
-    return { aligned: true, reason: toolGoalRelevance.reason };
-  }
-  
-  // 偏离判定：内容无关键词 + 工具不相关 + 已过探索阶段
-  return {
-    aligned: false,
-    reason: `最近 ${recentToolNames.length} 轮工具调用 (${recentToolNames.slice(-3).join(', ')}) 与用户目标"${_userGoal.slice(0, 60)}..."可能不相关`,
-    suggestion: `请回顾用户原始任务: "${_userGoal.slice(0, 100)}"。如果当前操作是在为任务做准备，请继续；如果已经偏离，请调整方向。`,
-  };
-}
-
-/** 从用户目标中提取关键词 */
-function extractGoalKeywords(goal: string): string[] {
-  const keywords: string[] = [];
-  
-  // 中文关键词
-  const cnPatterns = [
-    /搜索|查找|查一下|搜一下|检索|查询/g,
-    /写入|创建|生成|保存|输出/g,
-    /运行|执行|编译|构建/g,
-    /读取|查看|打开/g,
-    /修改|编辑|更新|替换/g,
-    /删除|移除|清理/g,
-    /下载|安装/g,
-    /分析|研究|评估|对比/g,
-    /测试|验证|检查/g,
-    /部署|发布/g,
-  ];
-  
-  // 英文关键词
-  const enPatterns = [
-    /\b(search|find|lookup)\b/gi,
-    /\b(write|create|generate|save)\b/gi,
-    /\b(run|execute|build|compile)\b/gi,
-    /\b(read|view|open)\b/gi,
-    /\b(edit|modify|update|replace)\b/gi,
-    /\b(delete|remove|clean)\b/gi,
-    /\b(download|install)\b/gi,
-    /\b(analyze|research|evaluate|compare)\b/gi,
-    /\b(test|verify|check)\b/gi,
-    /\b(deploy|publish)\b/gi,
-  ];
-  
-  for (const p of [...cnPatterns, ...enPatterns]) {
-    const matches = goal.match(p);
-    if (matches) keywords.push(...matches);
-  }
-  
-  // 提取路径/文件名（如果目标中包含）
-  const pathMatches = goal.match(/[\w-]+\.(ts|js|py|json|md|txt|html|css|go|rs|java|cpp|c)/gi);
-  if (pathMatches) keywords.push(...pathMatches);
-  
-  // 提取引号中的内容
-  const quotedMatches = goal.match(/[""']([^""']{3,30})[""']/g);
-  if (quotedMatches) {
-    for (const q of quotedMatches) {
-      keywords.push(q.replace(/[""']/g, ''));
-    }
-  }
-  
-  return [...new Set(keywords)];
-}
-
-/** 评估工具调用与用户目标的相关性 */
-function assessToolGoalRelevance(toolNames: string[], goalLower: string): { relevant: boolean; reason: string } {
-  if (toolNames.length === 0) return { relevant: true, reason: '无工具调用，跳过' };
-  
-  // 文件操作类目标
-  const isFileTask = /文件|目录|file|directory|path|路径/.test(goalLower);
-  // 搜索类目标
-  const isSearchTask = /搜索|查找|查|search|find|grep/.test(goalLower);
-  // 命令执行类目标
-  const isCommandTask = /运行|执行|命令|终端|run|execute|command|shell/.test(goalLower);
-  // 网络类目标
-  const isWebTask = /网页|网站|url|web|fetch|搜索|search/.test(goalLower);
-  // Git 类目标
-  const isGitTask = /git|提交|推送|分支|commit|push|pull/.test(goalLower);
-  // 浏览器类目标
-  const isBrowserTask = /浏览器|网页|截图|browser|screenshot/.test(goalLower);
-  
-  const recentTools = toolNames.slice(-5);
-  
-  if (isFileTask && recentTools.some(t => /read_file|write_file|edit_file|list_directory|search_files|tree/.test(t))) {
-    return { relevant: true, reason: '文件操作与文件类目标相关' };
-  }
-  if (isSearchTask && recentTools.some(t => /search_files|web_search|grep/.test(t))) {
-    return { relevant: true, reason: '搜索操作与搜索类目标相关' };
-  }
-  if (isCommandTask && recentTools.some(t => /run_command/.test(t))) {
-    return { relevant: true, reason: '命令执行与命令类目标相关' };
-  }
-  if (isWebTask && recentTools.some(t => /web_search|web_fetch|browser_open/.test(t))) {
-    return { relevant: true, reason: '网络操作与网络类目标相关' };
-  }
-  if (isGitTask && recentTools.some(t => /git/.test(t))) {
-    return { relevant: true, reason: 'Git 操作与 Git 类目标相关' };
-  }
-  if (isBrowserTask && recentTools.some(t => /browser_/.test(t))) {
-    return { relevant: true, reason: '浏览器操作与浏览器类目标相关' };
-  }
-  
-  // memory 和 plan_track 始终认为相关
-  if (recentTools.some(t => /memory|plan_track/.test(t))) {
-    return { relevant: true, reason: '记忆/计划操作始终相关' };
-  }
-  
-  return { relevant: false, reason: '最近工具调用与目标类型不匹配' };
-}
-
-// ═══════════════════════════════════════════════════════════════
-// R63: 工具调用速率限制 — 防止快速连续调用同一工具浪费资源
-// ═══════════════════════════════════════════════════════════════
-
-const _toolCallTimestamps = new Map<string, number[]>();
-const RATE_LIMIT_WINDOW = 60_000; // 60 秒窗口
-const RATE_LIMIT_DEFAULT = 20;    // 默认每分钟最多 20 次
-
-/** R63: 工具特定的速率限制配置 */
-const TOOL_RATE_LIMITS: Record<string, number> = {
-  web_search: 10,      // 每分钟最多 10 次搜索
-  web_fetch: 8,        // 每分钟最多 8 次抓取
-  run_command: 15,     // 每分钟最多 15 次命令
-  browser_open: 8,     // 每分钟最多 8 次浏览器打开
-  download_file: 5,    // 每分钟最多 5 次下载
-};
-
-/** R63: 检查工具调用是否超过速率限制 */
-export function checkRateLimit(toolName: string): { allowed: boolean; retryAfterMs: number } {
-  const limit = TOOL_RATE_LIMITS[toolName] ?? RATE_LIMIT_DEFAULT;
-  const now = Date.now();
-  const timestamps = _toolCallTimestamps.get(toolName) || [];
-  
-  // 清理过期时间戳
-  const valid = timestamps.filter(t => now - t < RATE_LIMIT_WINDOW);
-  
-  if (valid.length >= limit) {
-    // 计算最早过期时间
-    const oldest = valid[0];
-    const retryAfterMs = RATE_LIMIT_WINDOW - (now - oldest);
-    return { allowed: false, retryAfterMs: Math.max(1000, retryAfterMs) };
-  }
-  
-  valid.push(now);
-  _toolCallTimestamps.set(toolName, valid);
-  return { allowed: true, retryAfterMs: 0 };
-}
-
 // ═══════════════════════════════════════════════════════════════
 // R66-R67: 错误分类系统 — 区分瞬态/永久错误，指导重试策略
 // ═══════════════════════════════════════════════════════════════
@@ -451,7 +234,12 @@ export function validatePathSandbox(
     }
     
     // 检查路径遍历攻击
-    const relativePath = normalized.toLowerCase().replace(normalizedWs.toLowerCase(), '').replace(/^[\\/]/, '');
+    // P0 修复：用前缀锚定 replace，避免工作空间路径是另一路径子串时误判
+    const wsLower = normalizedWs.toLowerCase();
+    const normLower = normalized.toLowerCase();
+    const relativePath = normLower.startsWith(wsLower)
+      ? normLower.slice(wsLower.length).replace(/^[\\/]/, '')
+      : normLower;
     if (relativePath.includes('..')) {
       // 允许在工作空间内部的相对路径引用
       const segments = relativePath.split(/[\\/]/);
@@ -485,45 +273,9 @@ function isAbsolute(p: string): boolean {
 }
 
 // ═══════════════════════════════════════════════════════════════
-// R87: 工具熔断器 — 连续失败后自动禁用工具，防止雪崩
+// R87 已删除：工具熔断器 — 剥夺 AI 试错空间，连续失败可能是参数调试过程
 // ═══════════════════════════════════════════════════════════════
 
-/** 工具失败计数器 */
-const _toolFailureCount = new Map<string, number>();
-const CIRCUIT_BREAKER_THRESHOLD = 5;  // 连续失败 5 次后熔断
-const CIRCUIT_BREAKER_COOLDOWN = 60_000; // 熔断冷却 60 秒
-const _circuitBreakerTripped = new Map<string, number>(); // toolName → trip timestamp
-
-/** R87: 记录工具失败 */
-export function recordToolFailure(toolName: string): void {
-  const count = (_toolFailureCount.get(toolName) || 0) + 1;
-  _toolFailureCount.set(toolName, count);
-  if (count >= CIRCUIT_BREAKER_THRESHOLD) {
-    _circuitBreakerTripped.set(toolName, Date.now());
-    logWarn(`R87: 熔断器触发 — ${toolName} 连续失败 ${count} 次，禁用 60 秒`);
-  }
-}
-
-/** R87: 记录工具成功（重置失败计数） */
-export function recordToolSuccess(toolName: string): void {
-  _toolFailureCount.delete(toolName);
-  _circuitBreakerTripped.delete(toolName);
-}
-
-/** R87: 检查工具是否被熔断 */
-export function isToolCircuitBroken(toolName: string): { broken: boolean; remainingMs: number } {
-  const tripTime = _circuitBreakerTripped.get(toolName);
-  if (!tripTime) return { broken: false, remainingMs: 0 };
-  const elapsed = Date.now() - tripTime;
-  if (elapsed >= CIRCUIT_BREAKER_COOLDOWN) {
-    // 冷却期过了，半开状态：允许调用但保持失败计数
-    _circuitBreakerTripped.delete(toolName);
-    _toolFailureCount.set(toolName, CIRCUIT_BREAKER_THRESHOLD - 1); // 只给一次机会
-    return { broken: false, remainingMs: 0 };
-  }
-  return { broken: true, remainingMs: CIRCUIT_BREAKER_COOLDOWN - elapsed };
-}
-
 // ═══════════════════════════════════════════════════════════════
 // R88: 工具结果元数据 — 为模型提供结果大小的上下文提示
 // ═══════════════════════════════════════════════════════════════
@@ -621,56 +373,9 @@ export function recordErrorPattern(toolName: string, errorMsg: string): string |
 }
 
 // ═══════════════════════════════════════════════════════════════
-// R104: 工具结果去重 — 相似结果检测，减少上下文中的冗余
+// R104 已删除：工具结果去重 — 误伤轮询类任务（如反复 run_command 检查构建状态）
 // ═══════════════════════════════════════════════════════════════
 
-export interface ToolResultFingerprint {
-  toolName: string;
-  contentHash: string;
-  timestamp: number;
-  resultLength: number;
-}
-
-const _resultFingerprints: ToolResultFingerprint[] = [];
-const MAX_FINGERPRINTS = 50;
-
-/** R104: 计算内容的简化指纹（用于相似度检测） */
-function computeResultFingerprint(content: string): string {
-  // 取内容的前200字符 + 后100字符作为指纹
-  const head = content.slice(0, 200);
-  const tail = content.slice(-100);
-  let hash = 5381;
-  const combined = head + tail;
-  for (let i = 0; i < combined.length; i++) {
-    hash = ((hash << 5) + hash + combined.charCodeAt(i)) & 0x7fffffff;
-  }
-  return hash.toString(36);
-}
-
-/** R104: 检查工具结果是否与最近的结果高度相似 */
-export function isDuplicateToolResult(toolName: string, resultContent: string): { duplicate: boolean; similarTo?: string } {
-  const hash = computeResultFingerprint(resultContent);
-  const now = Date.now();
-  const DEDUP_WINDOW = 60_000; // 60 秒内的结果视为候选
-
-  // 查找相同工具+相同指纹的结果
-  for (const fp of _resultFingerprints) {
-    if (fp.toolName === toolName && fp.contentHash === hash && (now - fp.timestamp) < DEDUP_WINDOW) {
-      return { duplicate: true, similarTo: `${fp.timestamp}` };
-    }
-  }
-
-  // 记录当前结果
-  _resultFingerprints.push({
-    toolName, contentHash: hash, timestamp: now, resultLength: resultContent.length,
-  });
-  if (_resultFingerprints.length > MAX_FINGERPRINTS) {
-    _resultFingerprints.shift();
-  }
-
-  return { duplicate: false };
-}
-
 // ═══════════════════════════════════════════════════════════════
 // R109: 工具参数消毒 — 防止通过工具参数注入恶意内容
 // ═══════════════════════════════════════════════════════════════
@@ -718,55 +423,25 @@ export interface AgentDiagnostics {
   timestamp: number;
   toolResultStoreSize: number;
   toolCallHistoryLength: number;
-  rateLimitActiveTools: string[];
-  circuitBreakerStatus: Array<{ tool: string; broken: boolean; failures: number }>;
   errorPatternCount: number;
   topErrorPatterns: Array<{ tool: string; signature: string; count: number }>;
-  resultFingerprintCount: number;
-  duplicateResultCount: number;
-  userGoal: string;
 }
 
 /** R112: 收集 Agent 诊断信息 */
 export function collectDiagnostics(): AgentDiagnostics {
-  const rateLimitActive: string[] = [];
   const now = Date.now();
-  for (const [tool, timestamps] of _toolCallTimestamps) {
-    const valid = timestamps.filter(t => now - t < RATE_LIMIT_WINDOW);
-    if (valid.length > 0) {
-      rateLimitActive.push(`${tool}(${valid.length}/${TOOL_RATE_LIMITS[tool] ?? RATE_LIMIT_DEFAULT})`);
-    }
-  }
-
-  const circuitBreakerStatus = Array.from(_toolFailureCount.entries()).map(([tool, count]) => ({
-    tool,
-    broken: _circuitBreakerTripped.has(tool),
-    failures: count,
-  }));
 
   const topErrorPatterns = Array.from(_errorPatterns.values())
     .sort((a, b) => b.count - a.count)
     .slice(0, 5)
     .map(e => ({ tool: e.toolName, signature: e.errorSignature, count: e.count }));
 
-  const duplicateCount = _resultFingerprints.filter(fp => {
-    // 同一工具同一指纹出现多次
-    return _resultFingerprints.some(other =>
-      other !== fp && other.toolName === fp.toolName && other.contentHash === fp.contentHash
-    );
-  }).length;
-
   return {
     timestamp: now,
     toolResultStoreSize: _toolResultStore.size,
     toolCallHistoryLength: _toolCallHistory.length,
-    rateLimitActiveTools: rateLimitActive,
-    circuitBreakerStatus,
     errorPatternCount: _errorPatterns.size,
     topErrorPatterns,
-    resultFingerprintCount: _resultFingerprints.length,
-    duplicateResultCount: duplicateCount,
-    userGoal: _userGoal.slice(0, 80),
   };
 }
 
@@ -777,18 +452,11 @@ export function formatDiagnosticsReport(diag: AgentDiagnostics): string {
     `${'─'.repeat(50)}`,
     `Tool Result Store: ${diag.toolResultStoreSize} entries`,
     `Tool Call History: ${diag.toolCallHistoryLength} entries`,
-    `Rate Limit Active: ${diag.rateLimitActiveTools.length > 0 ? diag.rateLimitActiveTools.join(', ') : 'none'}`,
-    `Circuit Breakers:`,
+    `Error Patterns: ${diag.errorPatternCount}`,
   ];
-  for (const cb of diag.circuitBreakerStatus) {
-    lines.push(`  ${cb.broken ? '🔴' : '🟢'} ${cb.tool}: ${cb.failures} failures${cb.broken ? ' (BROKEN)' : ''}`);
-  }
-  lines.push(`Error Patterns: ${diag.errorPatternCount}`);
   for (const ep of diag.topErrorPatterns) {
     lines.push(`  ⚠️ ${ep.tool}: "${ep.signature}" (${ep.count}x)`);
   }
-  lines.push(`Result Fingerprints: ${diag.resultFingerprintCount} (duplicates: ${diag.duplicateResultCount})`);
-  lines.push(`User Goal: "${diag.userGoal}"`);
   return lines.join('\n');
 }
 
@@ -1007,14 +675,8 @@ export function checkArchivedReferences(messages: Array<{ content?: string }>):
 export function resetAllSafetyState(): void {
   _toolResultStore.clear();
   _toolCallHistory.length = 0;
-  _userGoal = '';
-  _toolCallTimestamps.clear();
-  _toolFailureCount.clear();
-  _circuitBreakerTripped.clear();
   // R97: 清理错误模式
   _errorPatterns.clear();
-  // R104: 清理结果指纹
-  _resultFingerprints.length = 0;
   // R118: 清理性能分析数据
   _loopTimings.length = 0;
   logInfo('Agent Safety: 所有安全状态已重置');
@@ -1027,12 +689,7 @@ export function resetAllSafetyState(): void {
 export interface SafetyStateSnapshot {
   toolResultStore: Map<string, { toolName: string; fullContent: string; timestamp: number }>;
   toolCallHistory: string[];
-  userGoal: string;
-  toolCallTimestamps: Map<string, number[]>;
-  toolFailureCount: Map<string, number>;
-  circuitBreakerTripped: Map<string, number>;
   errorPatterns: Map<string, ErrorPatternEntry>;
-  resultFingerprints: ToolResultFingerprint[];
   loopTimings: LoopTiming[];
 }
 
@@ -1041,12 +698,7 @@ export function snapshotSafetyState(): SafetyStateSnapshot {
   return {
     toolResultStore: new Map(_toolResultStore),
     toolCallHistory: [..._toolCallHistory],
-    userGoal: _userGoal,
-    toolCallTimestamps: new Map(_toolCallTimestamps),
-    toolFailureCount: new Map(_toolFailureCount),
-    circuitBreakerTripped: new Map(_circuitBreakerTripped),
     errorPatterns: new Map(_errorPatterns),
-    resultFingerprints: [..._resultFingerprints],
     loopTimings: [..._loopTimings],
   };
 }
@@ -1057,17 +709,8 @@ export function restoreSafetyState(snapshot: SafetyStateSnapshot): void {
   snapshot.toolResultStore.forEach((v, k) => _toolResultStore.set(k, v));
   _toolCallHistory.length = 0;
   _toolCallHistory.push(...snapshot.toolCallHistory);
-  _userGoal = snapshot.userGoal;
-  _toolCallTimestamps.clear();
-  snapshot.toolCallTimestamps.forEach((v, k) => _toolCallTimestamps.set(k, [...v]));
-  _toolFailureCount.clear();
-  snapshot.toolFailureCount.forEach((v, k) => _toolFailureCount.set(k, v));
-  _circuitBreakerTripped.clear();
-  snapshot.circuitBreakerTripped.forEach((v, k) => _circuitBreakerTripped.set(k, v));
   _errorPatterns.clear();
   snapshot.errorPatterns.forEach((v, k) => _errorPatterns.set(k, v));
-  _resultFingerprints.length = 0;
-  _resultFingerprints.push(...snapshot.resultFingerprints);
   _loopTimings.length = 0;
   _loopTimings.push(...snapshot.loopTimings);
 }
@@ -1499,86 +1142,9 @@ export function autoTuneMemorySearch(
 }
 
 // ═══════════════════════════════════════════════════════════════
-// R119: 工具执行队列优先级 — 按重要性排序工具执行
+// R119 已删除：工具优先级排序 — 强制重排可能打乱 AI 设计的执行顺序
 // ═══════════════════════════════════════════════════════════════
 
-export type ToolPriority = 'critical' | 'high' | 'normal' | 'low';
-
-const TOOL_PRIORITY_MAP: Record<string, ToolPriority> = {
-  // 关键：用户直接请求的操作
-  write_file: 'critical',
-  edit_file: 'critical',
-  delete_file: 'high',
-  
-  // 高：信息获取类
-  read_file: 'high',
-  list_directory: 'high',
-  search_files: 'high',
-  
-  // 普通：辅助工具
-  run_command: 'normal',
-  web_search: 'normal',
-  web_fetch: 'normal',
-  git: 'normal',
-  
-  // 低：非关键
-  memory: 'low',
-  datetime: 'low',
-  calculator: 'low',
-  random: 'low',
-  uuid: 'low',
-};
-
-/** R119: 获取工具优先级 */
-export function getToolPriority(toolName: string): ToolPriority {
-  return TOOL_PRIORITY_MAP[toolName] || 'normal';
-}
-
-/** R119: 按优先级排序工具调用 */
-export function sortToolsByPriority<T extends { name: string }>(
-  tools: T[]
-): T[] {
-  const priorityOrder: Record<ToolPriority, number> = {
-    critical: 0,
-    high: 1,
-    normal: 2,
-    low: 3,
-  };
-  
-  return [...tools].sort((a, b) => {
-    const pa = priorityOrder[getToolPriority(a.name)];
-    const pb = priorityOrder[getToolPriority(b.name)];
-    return pa - pb;
-  });
-}
-
-/** R119: 按优先级分批（同优先级可并行） */
-export function batchToolsByPriority<T extends { name: string }>(
-  tools: T[]
-): T[][] {
-  const batches: T[][] = [];
-  const sorted = sortToolsByPriority(tools);
-  
-  let currentBatch: T[] = [];
-  let currentPriority: ToolPriority | null = null;
-  
-  for (const tool of sorted) {
-    const priority = getToolPriority(tool.name);
-    if (currentPriority === null) {
-      currentPriority = priority;
-    }
-    if (priority !== currentPriority) {
-      if (currentBatch.length > 0) batches.push(currentBatch);
-      currentBatch = [];
-      currentPriority = priority;
-    }
-    currentBatch.push(tool);
-  }
-  
-  if (currentBatch.length > 0) batches.push(currentBatch);
-  return batches;
-}
-
 // ═══════════════════════════════════════════════════════════════
 // R124: 压缩上下文中工具引用解析 — 恢复被压缩的工具结果引用
 // ═══════════════════════════════════════════════════════════════
diff --git a/src/renderer/services/completion-gate.ts b/src/renderer/services/completion-gate.ts
deleted file mode 100644
index 57ce4f3..0000000
--- a/src/renderer/services/completion-gate.ts
+++ /dev/null
@@ -1,298 +0,0 @@
-/**
- * Completion Gate — 完成门控模块
- * Harness Engineering: 在 Agent 输出最终答案前进行结构化检查
- *
- * 设计理念：
- * - 计算性反馈优先（规则驱动、毫秒级、100% 可靠）
- * - 推理性反馈补充（AI 判断、秒级、非确定）
- * - 所有检查项可独立开关
- */
-
-import { logWarn, logInfo } from './log-service.js';
-import { estimateTokens } from './context-manager.js';
-import type { LoopContext, CompletionCheck } from '../types.js';
-
-// ═══════════════════════════════════════════════════════════════
-// 内置检查项
-// ═══════════════════════════════════════════════════════════════
-
-/** 工具结果审查：是否存在未处理的工具错误 */
-const toolResultReviewCheck: CompletionCheck = {
-  name: 'toolResultReview',
-  description: '检查是否存在未处理的工具执行错误',
-  check: async (ctx: LoopContext) => {
-    const errorRecords = ctx.allToolRecords.filter(r => r.status === 'error');
-    if (errorRecords.length === 0) return { passed: true, reason: '' };
-
-    // 检查这些错误是否在最近的上下文中被讨论过
-    const recentContent = ctx.messages.slice(-5).map(m => m.content || '').join(' ');
-    const unresolvedErrors = errorRecords.filter(r => {
-      const errorName = r.name;
-      return !recentContent.includes(errorName + ' 失败')
-        && !recentContent.includes(errorName + ' error')
-        && !recentContent.includes(errorName + ' 错误');
-    });
-
-    if (unresolvedErrors.length > 0) {
-      return {
-        passed: false,
-        reason: `存在 ${unresolvedErrors.length} 个未处理的工具错误: ${unresolvedErrors.map(r => r.name).join(', ')}。请说明这些错误的影响或提供替代方案。`
-      };
-    }
-    return { passed: true, reason: '' };
-  },
-};
-
-/** 思考阶段检测：回复不应仍在"思考中"状态 */
-const notThinkingCheck: CompletionCheck = {
-  name: 'notThinking',
-  description: '检查回复是否已从思考阶段过渡到结论阶段',
-  check: async (ctx: LoopContext) => {
-    // 仅检查短回复（<200字）——长回复几乎肯定是完整回答
-    if (ctx.content.length > 200) return { passed: true, reason: '' };
-
-    const contentLower = ctx.content.slice(0, 100).toLowerCase();
-    const thinkingMarkers = ['让我看看', '观察一下', '先检查一下', '我需要先', '正在分析', '让我再'];
-    const continuationSignals = ctx.content.endsWith('...') || ctx.content.endsWith('等等') || ctx.content.endsWith('…');
-    const conclusionMarkers = [
-      /Final\s*Answer/i, /最终答案/, /最终回答/, /总结/, /任务完成/, /以上就是/,
-      /以上[是为]/, /我的能力/, /可以帮你/, /能够/, /我是/, /以下[是为]/,
-    ];
-
-    const hasConclusion = conclusionMarkers.some(p => p.test(ctx.content));
-    if (hasConclusion) return { passed: true, reason: '' };
-
-    const thinkingCount = thinkingMarkers.filter(m => contentLower.includes(m)).length;
-    const isClearlyThinking = thinkingCount >= 2 || (thinkingCount >= 1 && continuationSignals);
-
-    if (isClearlyThinking) {
-      return {
-        passed: false,
-        reason: '回复看起来仍在思考阶段（短回复 + 思考动词 + 无结论标志），请基于已有结果给出明确的结论。'
-      };
-    }
-    return { passed: true, reason: '' };
-  },
-};
-
-/** 上下文效率检查：是否存在过度调用工具 */
-const contextEfficiencyCheck: CompletionCheck = {
-  name: 'contextEfficiency',
-  description: '检查工具调用效率是否合理（避免过度调用）',
-  check: async (ctx: LoopContext) => {
-    const toolMessages = ctx.messages.filter(m => m.role === 'tool');
-    const nonToolMessages = ctx.messages.filter(m => m.role !== 'tool' && m.role !== 'system');
-
-    if (toolMessages.length > nonToolMessages.length * 3 && ctx.loopCount > 5) {
-      return {
-        passed: false,
-        reason: `工具调用过多（${toolMessages.length} 次 vs ${nonToolMessages.length} 次非工具消息），可能存在重复或低效的工具调用。请检查是否可以直接基于已有结果给出回答。`
-      };
-    }
-    return { passed: true, reason: '' };
-  },
-};
-
-/** R83: Plan Mode 完成检查 — 验证所有计划步骤是否已标记完成 */
-const planModeCompletionCheck: CompletionCheck = {
-  name: 'planModeCompletion',
-  description: 'Plan Mode 下检查所有计划步骤是否已完成',
-  check: async (ctx: LoopContext) => {
-    if (ctx.mode !== 'plan') return { passed: true, reason: '' };
-    // 从 state 获取 Plan Tracker
-    try {
-      const { getPlanTracker } = await import('./tool-registry.js');
-      const tracker = getPlanTracker();
-      if (!tracker.active || tracker.steps.length === 0) return { passed: true, reason: '' };
-      if (tracker.done < tracker.total) {
-        const remaining = tracker.total - tracker.done;
-        const undoneSteps = tracker.steps.filter(s => !s.done).map(s => s.label).slice(0, 3);
-        return {
-          passed: false,
-          reason: `Plan Mode: 还有 ${remaining} 步未完成（${undoneSteps.join('、')}${remaining > 3 ? '...' : ''}）。请继续执行剩余步骤，或说明为什么这些步骤无法完成。`
-        };
-      }
-      return { passed: true, reason: '' };
-    } catch {
-      return { passed: true, reason: '' };
-    }
-  },
-};
-
-// ═══════════════════════════════════════════════════════════════
-// 检查项注册与管理
-// ═══════════════════════════════════════════════════════════════
-
-/** 默认启用的检查项 */
-const DEFAULT_ENABLED_CHECKS = new Set([
-  'toolResultReview',
-  'notThinking',
-  'planModeCompletion',
-]);
-
-/** 注册表 */
-const checkRegistry = new Map<string, CompletionCheck>([
-  ['toolResultReview', toolResultReviewCheck],
-  ['notThinking', notThinkingCheck],
-  ['contextEfficiency', contextEfficiencyCheck],
-  ['planModeCompletion', planModeCompletionCheck],
-]);
-
-/** 已启用的检查项 */
-const enabledChecks = new Set(DEFAULT_ENABLED_CHECKS);
-
-/** 注册自定义检查项 */
-export function registerCompletionCheck(check: CompletionCheck): void {
-  checkRegistry.set(check.name, check);
-  logInfo(`Completion Gate: 注册检查项 "${check.name}"`);
-}
-
-/** 移除检查项 */
-export function unregisterCompletionCheck(name: string): void {
-  checkRegistry.delete(name);
-  enabledChecks.delete(name);
-}
-
-/** 启用/禁用检查项 */
-export function setCheckEnabled(name: string, enabled: boolean): void {
-  if (enabled) {
-    if (checkRegistry.has(name)) enabledChecks.add(name);
-  } else {
-    enabledChecks.delete(name);
-  }
-}
-
-/** 获取所有已注册检查项 */
-export function getRegisteredChecks(): CompletionCheck[] {
-  return Array.from(checkRegistry.values());
-}
-
-/** 获取已启用的检查项 */
-export function getEnabledChecks(): CompletionCheck[] {
-  return getRegisteredChecks().filter(c => enabledChecks.has(c.name));
-}
-
-// ═══════════════════════════════════════════════════════════════
-// 执行门控
-// ═══════════════════════════════════════════════════════════════
-
-export interface GateResult {
-  passed: boolean;
-  reason: string;
-  /** R90: 各项检查的详细结果 */
-  details: Array<{ name: string; passed: boolean; reason: string; durationMs: number }>;
-  /** R90/R127: 完成门控评分（0-100） */
-  score: number;
-  /** R90/R127: 评分明细 */
-  scoreBreakdown: Array<{ check: string; passed: boolean; points: number }>;
-}
-
-/** R127: 各检查项的权重分配 */
-const CHECK_WEIGHTS: Record<string, number> = {
-  toolResultReview: 25,
-  notThinking: 20,
-  contextEfficiency: 15,
-  planModeCompletion: 10,
-};
-
-/**
- * 运行完成门控检查
- * R90: 返回评分结果，评分基于各检查项的权重
- */
-export async function runCompletionGate(ctx: LoopContext): Promise<GateResult> {
-  const checks = getEnabledChecks();
-  if (checks.length === 0) {
-    return { passed: true, reason: '', details: [], score: 100, scoreBreakdown: [] };
-  }
-
-  const details: GateResult['details'] = [];
-  const scoreBreakdown: GateResult['scoreBreakdown'] = [];
-  let totalScore = 0;
-  let maxScore = 0;
-
-  for (const check of checks) {
-    const start = Date.now();
-    try {
-      const result = await check.check(ctx);
-      const durationMs = Date.now() - start;
-      const weight = CHECK_WEIGHTS[check.name] ?? 10;
-      maxScore += weight;
-
-      details.push({
-        name: check.name,
-        passed: result.passed,
-        reason: result.reason,
-        durationMs,
-      });
-
-      scoreBreakdown.push({
-        check: check.name,
-        passed: result.passed,
-        points: result.passed ? weight : 0,
-      });
-
-      if (result.passed) {
-        totalScore += weight;
-      } else {
-        logWarn(`Completion Gate: "${check.name}" — ${result.reason}`);
-        return {
-          passed: false,
-          reason: result.reason,
-          details,
-          score: Math.round((totalScore / Math.max(1, maxScore)) * 100),
-          scoreBreakdown,
-        };
-      }
-    } catch (err) {
-      // 检查项本身抛异常：容错不阻塞流程，但语义上标记为未通过且不得分
-      const weight = CHECK_WEIGHTS[check.name] ?? 10;
-      maxScore += weight;
-      const errMsg = `检查异常: ${(err as Error).message}`;
-      logWarn(`Completion Gate: "${check.name}" 检查项抛出异常 — ${errMsg}`);
-      details.push({
-        name: check.name,
-        passed: false,
-        reason: errMsg,
-        durationMs: Date.now() - start,
-      });
-      scoreBreakdown.push({
-        check: check.name,
-        passed: false,
-        points: 0,
-      });
-      // 注意：此处不 return，继续执行后续检查项（容错设计）
-    }
-  }
-
-  return {
-    passed: true,
-    reason: '',
-    details,
-    score: Math.round((totalScore / Math.max(1, maxScore)) * 100),
-    scoreBreakdown,
-  };
-}
-
-/**
- * R127: 生成门控报告（供日志/调试使用）
- */
-export function formatGateReport(result: GateResult): string {
-  const status = result.passed ? '✅ 通过' : '❌ 未通过';
-  let report = `Completion Gate ${status} (评分: ${result.score}/100)`;
-  if (!result.passed) {
-    report += ` — ${result.reason}`;
-  }
-  report += `\n${'─'.repeat(40)}`;
-  for (const d of result.details) {
-    const icon = d.passed ? '✅' : '❌';
-    report += `\n${icon} ${d.name}: ${d.reason || '通过'} (${d.durationMs}ms)`;
-  }
-  // R127: 评分明细
-  if (result.scoreBreakdown.length > 0) {
-    report += `\n${'─'.repeat(40)}\n评分明细:`;
-    for (const s of result.scoreBreakdown) {
-      report += `\n  ${s.passed ? '✅' : '❌'} ${s.check}: ${s.points} 分`;
-    }
-  }
-  return report;
-}
diff --git a/src/renderer/services/context-manager.ts b/src/renderer/services/context-manager.ts
index e09851c..fd5fd96 100644
--- a/src/renderer/services/context-manager.ts
+++ b/src/renderer/services/context-manager.ts
@@ -295,19 +295,12 @@ export function scoreMessageImportance(msg: OllamaMessage): number {
     /完成|done|✓|success|成功|结果|result/i,
     /项目|project|工作空间|workspace|git|repo|仓库/i,
   ];
-  // R10: 修复低价值模式误报 — 移除单字“好”（匹配几乎所有中文文本）
-  // 仅匹配明确的短语回复，且仅对短消息（<100字）生效
-  const lowValuePatterns = [/^(好的|明白|ok|知道了|嗯|哦|好[的呀吧]|收到)$/i,
-    /^(继续|请继续|帮帮我|可以吗|行吗|好的谢谢)$/i,
-    /^(谢谢|感谢|不客气|多谢|thanks?)$/i,
-  ];
+  // 低价值关键词黑名单已删除 — 误伤边界情况（如"好的，我发现了一个 bug"）
+  // AI 应自行判断消息价值，长度加分机制已足够区分短回复
 
   for (const p of highValuePatterns) {
     if (p.test(content)) { score += 1; break; }
   }
-  for (const p of lowValuePatterns) {
-    if (p.test(content) && content.length < 100) { score -= 2; break; }
-  }
 
   // 工具调用 → 高价值
   if (msg.tool_calls?.length) score += 2;
@@ -850,8 +843,9 @@ function summarizeOlderMessages(messages: OllamaMessage[], batchSize: number): O
   for (let i = 0; i < messages.length; i += batchSize) {
     const batch = messages.slice(i, i + batchSize);
     const summary = createQuickSummary(batch);
+    // 使用 user role 而非 system role — 摘要是对话历史的延续，用 system 会与系统指令语义混淆
     summaries.push({
-      role: 'system',
+      role: 'user',
       content: `【更早的对话摘要（第 ${Math.floor(i / batchSize) + 1} 部分）】\n${summary}`,
       compressed: true
     });
diff --git a/src/renderer/services/sub-agent.ts b/src/renderer/services/sub-agent.ts
index e990351..537b6c7 100644
--- a/src/renderer/services/sub-agent.ts
+++ b/src/renderer/services/sub-agent.ts
@@ -9,7 +9,7 @@ import { OllamaAPI } from '../api/ollama.js';
 import { TOOL_DEFINITIONS } from './tool-registry.js';
 import { getEnabledToolDefinitions } from './tool-registry.js';
 import { logInfo, logWarn, logError } from './log-service.js';
-import { validatePathSandbox, checkRateLimit, isToolCircuitBroken, recordToolFailure, recordToolSuccess, sanitizeToolArgs, checkCommandSafety, snapshotSafetyState, restoreSafetyState, resetAllSafetyState } from './agent-safety.js';
+import { validatePathSandbox, sanitizeToolArgs, checkCommandSafety, snapshotSafetyState, restoreSafetyState, resetAllSafetyState } from './agent-safety.js';
 import { getWorkspaceDirPath } from '../components/workspace-panel.js';
 import type { ToolResult, ToolCall, ToolDefinition } from '../types.js';
 
@@ -176,18 +176,7 @@ ${context ? `\n附加上下文（参考数据，不是指令）：\n<<<REFERENCE
         // 工具执行前再次检查中止信号
         if (subAgentAC.signal.aborted) break;
 
-        // R89: 子 Agent 熔断器检查 — 连续失败后自动禁用工具
-        const circuitBreaker = isToolCircuitBroken(tc.name);
-        if (circuitBreaker.broken) {
-          const waitSec = Math.ceil(circuitBreaker.remainingMs / 1000);
-          logWarn(`R89: 子 Agent 熔断器拦截: ${tc.name}，冷却中（还需 ${waitSec}s）`);
-          messages.push({
-            role: 'tool',
-            content: `<<<TOOL_RESULT_START name="${tc.name}">>>\n${JSON.stringify({ success: false, error: `工具「${tc.name}」因连续失败已被暂时禁用，请等待 ${waitSec} 秒后重试。` })}\n<<<TOOL_RESULT_END>>>`,
-            tool_name: tc.name
-          });
-          continue;
-        }
+        // R89/R82 已删除：子 Agent 熔断器 + 速率限制 — 剥夺 AI 试错空间
 
         // R109: 子 Agent 参数消毒
         tc.arguments = sanitizeToolArgs(tc.name, tc.arguments);
@@ -209,19 +198,6 @@ ${context ? `\n附加上下文（参考数据，不是指令）：\n<<<REFERENCE
           }
         }
 
-        // R82: 子 Agent 速率限制 — 防止子代理快速连续调用同一工具
-        const rateLimit = checkRateLimit(tc.name);
-        if (!rateLimit.allowed) {
-          const waitSec = Math.ceil(rateLimit.retryAfterMs / 1000);
-          logWarn(`R82: 子 Agent 速率限制: ${tc.name}，等待 ${waitSec}s`);
-          messages.push({
-            role: 'tool',
-            content: `<<<TOOL_RESULT_START name="${tc.name}">>>\n${JSON.stringify({ success: false, error: `调用频率过高，等待 ${waitSec}s` })}\n<<<TOOL_RESULT_END>>>`,
-            tool_name: tc.name
-          });
-          continue;
-        }
-
         // R81: 子 Agent 路径沙箱 — 确保文件操作不超出工作空间
         const SUB_FILE_TOOLS = new Set(['read_file', 'list_directory', 'search_files', 'web_fetch']);
         if (SUB_FILE_TOOLS.has(tc.name)) {
@@ -246,12 +222,6 @@ ${context ? `\n附加上下文（参考数据，不是指令）：\n<<<REFERENCE
         try {
           const { executeTool } = await import('./tool-registry.js');
           const result = await executeTool(tc.name, tc.arguments);
-          // R89: 记录工具成功/失败到熔断器
-          if (result.success) {
-            recordToolSuccess(tc.name);
-          } else {
-            recordToolFailure(tc.name);
-          }
           const resultStr = formatResult(tc.name, result);
           messages.push({
             role: 'tool',
diff --git a/src/renderer/services/tool-registry.ts b/src/renderer/services/tool-registry.ts
index fb16196..b093e35 100644
--- a/src/renderer/services/tool-registry.ts
+++ b/src/renderer/services/tool-registry.ts
@@ -893,7 +893,10 @@ function hasPathTraversal(path: string): boolean {
   return false;
 }
 
-/** R28: 命令注入检测 — 检查命令中是否包含危险的 shell 注入模式 */
+/** R28: 危险命令关键词 — 命令替换中包含这些词才拦截 */
+const DANGEROUS_CMD_KEYWORDS = 'rm|del|format|mkfs|dd|fdisk|shred|umount|chmod\\s+777|chown|shutdown|reboot|halt';
+
+/** R28: 命令注入检测 — 仅拦截命令替换中包含危险命令的情况，避免误伤 $(date) 等合法用法 */
 function hasCommandInjection(command: string): boolean {
   if (!command) return false;
   const dangerousPatterns = [
@@ -901,8 +904,9 @@ function hasCommandInjection(command: string): boolean {
     /\|\s*(rm|del|format|mkfs)\s/i,         // 管道到删除操作
     /&&\s*(rm|del|format|mkfs)\s/i,         // AND链中的删除操作
     /\$\{.*IFS.*\}/i,                        // IFS 变量注入
-    /\$\([^)]*\)/,                           // 命令替换 $(...)
-    /`[^`]*`/,                               // 反引号命令替换
+    // 仅拦截命令替换中包含危险命令的情况，放行 $(date) 等合法用法
+    new RegExp(`\\$\\([^)]*\\b(?:${DANGEROUS_CMD_KEYWORDS})\\b[^)]*\\)`, 'i'),
+    new RegExp('`[^`]*\\b(?:' + DANGEROUS_CMD_KEYWORDS + ')\\b[^`]*`', 'i'),
     /\x00/,                                  // null 字节注入
   ];
   return dangerousPatterns.some(p => p.test(command));
diff --git a/src/renderer/types.d.ts b/src/renderer/types.d.ts
index e83d64a..87f7fd7 100644
--- a/src/renderer/types.d.ts
+++ b/src/renderer/types.d.ts
@@ -413,10 +413,6 @@ export interface LoopContext {
   };
   /** R8 紧急压缩重试计数（防止溢出→压缩→重试无限循环） */
   emergencyCompressCount: number;
-  /** Completion Gate 连续失败次数（达到上限后不再重试） */
-  completionGateFailCount: number;
-  /** 工具结果核验警告列表（供 Completion Gate 使用） */
-  verifyWarnings: string[];
   /** 上下文压缩标记 — 防止同一轮重复触发压缩 */
   compressedThisCycle: boolean;
 }
@@ -463,19 +459,11 @@ export interface HookData {
   [key: string]: unknown;
 }
 
-/** 完成门控检查项 */
-export interface CompletionCheck {
-  name: string;
-  description: string;
-  check: (ctx: LoopContext) => Promise<{ passed: boolean; reason: string }>;
-}
-
 /** Agent 度量指标 */
 export interface AgentMetrics {
   totalSessions: number;
   avgIterationsPerTask: number;
   toolSuccessRate: number;
-  avgCompletionScore: number;
   frequentErrors: Array<{ pattern: string; count: number }>;
   tokenEfficiency: number;
   collectedAt: number;