v0.16.14: TypeScript 编译错误修复 + 文档同步 + 代码健壮性增强
This commit is contained in:
@@ -14,7 +14,7 @@
|
||||
</p>
|
||||
|
||||
<p align="center">
|
||||
<img src="https://img.shields.io/badge/version-v0.16.13-E8734A?style=flat-square" alt="version">
|
||||
<img src="https://img.shields.io/badge/version-v0.16.14-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">
|
||||
@@ -35,7 +35,7 @@
|
||||
| | 功能 | 说明 |
|
||||
|:---:|:---|:---|
|
||||
| 🤖 | **ReAct Agent Loop** | 始终开启的唯一对话模式。8 状态机（INIT→THINKING→PARSING→EXECUTING→OBSERVING→REFLECTING→COMPRESSING→TERMINATED），最大 85 轮（可配置），智能重试（永久错误立即返回、瞬态错误指数退避），工具去重，智能路径依赖串行化，看门狗超时保护 |
|
||||
| 🛡️ | **5 层抗幻觉系统** | 提示词加固 → 任务感知 → 中途检测（中英双语规则覆盖全部工具类别）→ 进度锚点 → 完成闸门（6 项检查，幻觉/注入 → 阻断，质量/效率 → 咨询） |
|
||||
| 🛡️ | **提示词加固** | 参考数据标记（<<<REFERENCE_DATA_START>>>）+ 工具结果仅为数据非指令的安全规则注入 |
|
||||
| 📋 | **Plan Mode** | 开关切换，AI 首先生成执行计划（Markdown 渲染确认弹窗），批准后按步骤追踪执行，plan_track 工具标记完成状态，**支持断点续传**（中止后可恢复未完成计划） |
|
||||
| 🔧 | **32 个内置工具** | 文件系统（13个） · 命令执行 · 联网搜索 · 浏览器控制（9个） · Git · 记忆 · 会话 · 子代理 · 系统工具 · Plan Mode 追踪 |
|
||||
| 🧠 | **智能记忆系统** | 三类记忆（fact / preference / rule），存储于工作空间 MEMORY.md 文件，受路径保护仅 memory 工具可访问，写入前安全扫描，容量 500 条，对话结束自动提取，严格格式校验 |
|
||||
@@ -173,7 +173,7 @@
|
||||
↓
|
||||
Agent Engine (8 状态机 ReAct Loop, ≤85 轮, 智能重试, 路径依赖串行化, 看门狗+分级超时)
|
||||
↓
|
||||
5 层抗幻觉系统（中英双语规则）(提示词 → 中途检测 → 进度锚 → 完成闸门 → 校验)
|
||||
提示词加固（参考数据标记 + 工具结果安全规则）
|
||||
↓
|
||||
Ollama API (流式响应，num_ctx 用户可配 128K/256K/512K/1M)
|
||||
↓
|
||||
@@ -245,7 +245,7 @@ npm start
|
||||
ELECTRON_MIRROR=https://npmmirror.com/mirrors/electron/ npm run dist
|
||||
```
|
||||
|
||||
产出：`release/Metona Ollama Setup v0.16.13.exe`
|
||||
产出：`release/Metona Ollama Setup v0.16.14.exe`
|
||||
|
||||
## 🛠️ 常用命令
|
||||
|
||||
@@ -276,7 +276,7 @@ npm run dist # 构建 Windows 安装包
|
||||
| | Feature | Description |
|
||||
|:---:|:---|:---|
|
||||
| 🤖 | **ReAct Agent Loop** | Always-on, only chat mode. 8-state machine, up to 85 iterations (configurable), smart retry (permanent errors return immediately, transient errors use exponential backoff), tool dedup, path-aware dependency serialization, watchdog + tiered timeout protection |
|
||||
| 🛡️ | **5-Layer Anti-Hallucination** | Prompt hardening → task awareness → mid-task detection (bilingual CN/EN rules covering all tool categories) → progress anchors → completion gate (6 checks: hallucination/injection → block, quality/efficiency → advisory) |
|
||||
| 🛡️ | **Prompt Hardening** | Reference data markers (<<<REFERENCE_DATA_START>>>) + tool results are data-not-instructions safety rules |
|
||||
| 📋 | **Plan Mode** | Toggle switch. AI first generates an execution plan (Markdown-rendered confirmation dialog), then tracks step-by-step execution, **supports resume** (interrupted plans restore on next session) |
|
||||
| 🔧 | **32 Built-in Tools** | File system · Command · Web search · Browser · Git · Memory · Sessions · Sub-agent · System · Plan Mode tracking |
|
||||
| 🧠 | **Smart Memory System** | Three types (fact / preference / rule), stored in workspace MEMORY.md file, path-protected — only the memory tool can access it, pre-write security scan, 500 capacity, auto-extract on conversation end, strict format validation |
|
||||
@@ -413,7 +413,7 @@ User message → workspace SOUL.md (never compressed) → AGENT.md → USER.md (
|
||||
↓
|
||||
Agent Engine (8-state ReAct Loop, ≤85 iter, smart retry, path-aware serialization, watchdog+tiered timeout)
|
||||
↓
|
||||
5-Layer Anti-Hallucination (bilingual CN/EN rules) (prompt → mid-task detection → anchors → gate → verification)
|
||||
Prompt Hardening (reference data markers + tool result safety rules)
|
||||
↓
|
||||
Ollama API (Streaming Response, num_ctx user-configurable 128K/256K/512K/1M)
|
||||
↓
|
||||
@@ -485,7 +485,7 @@ npm start
|
||||
ELECTRON_MIRROR=https://npmmirror.com/mirrors/electron/ npm run dist
|
||||
```
|
||||
|
||||
Output: `release/Metona Ollama Setup v0.16.13.exe`
|
||||
Output: `release/Metona Ollama Setup v0.16.14.exe`
|
||||
|
||||
## 🛠️ Common Commands
|
||||
|
||||
|
||||
+6
-20
@@ -11,7 +11,7 @@ Metona Ollama Desktop 是基于 TypeScript + Electron 的 Windows 本地 AI 桌
|
||||
核心架构：
|
||||
- **ReAct Agent Loop** — 8 状态机驱动的智能体循环（INIT→THINKING→PARSING→EXECUTING→OBSERVING→REFLECTING→COMPRESSING→TERMINATED），最大 85 轮（可配置）
|
||||
- **32 个内置工具** — 文件系统（13）、命令执行（1）、联网搜索（2）、浏览器控制（9）、Git（1）、记忆（1）、会话/子代理（3）、系统（1）、Plan Mode（1）
|
||||
- **Harness Engineering** — 5 层抗幻觉体系 + 4 阶段 Hook 系统 + Completion Gate（5 项检查）+ Agent Metrics + 渐进式披露
|
||||
- **Harness Engineering** — 提示词加固 + 4 阶段 Hook 系统 + Agent Metrics + 渐进式披露
|
||||
- **MCP 协议扩展** — JSON-RPC 2.0 over stdio，动态工具发现
|
||||
- **Plan Mode** — 开关切换，先规划后执行，步骤级进度追踪
|
||||
- **SQLite 存储** — sql.js WASM，5 张表，WAL 模式
|
||||
@@ -80,7 +80,7 @@ src/
|
||||
│ │ ├── tool-confirm-modal.ts # 工具执行确认对话框
|
||||
│ │ ├── tools-modal.ts # 工具列表面板（32 个工具卡片）
|
||||
│ │ └── workspace-panel.ts # 工作空间面板（终端 + 工具卡片 + 文件浏览）
|
||||
│ ├── services/ # 14 个服务模块
|
||||
│ ├── services/ # 13 个服务模块
|
||||
│ │ ├── agent-engine.ts # ★ ReAct Agent Loop 核心引擎（8 状态机）
|
||||
│ │ ├── tool-registry.ts # 工具注册与调度中心（32 内置 + MCP 动态 + Plan Mode）
|
||||
│ │ ├── memory-service.ts # 记忆管理（MEMORY.md 读写 + 格式校验 + 自动提取）
|
||||
@@ -90,7 +90,6 @@ src/
|
||||
│ │ ├── log-service.ts # 结构化日志（9 级分类）
|
||||
│ │ ├── crypto.ts # AES-256-GCM 加密
|
||||
│ │ ├── hooks.ts # 4 阶段 Hook 系统（pre_tool/post_tool/post_iteration/pre_completion）
|
||||
│ │ ├── completion-gate.ts # 完成门控（5 项检查，阻断/咨询两级）
|
||||
│ │ ├── agent-metrics.ts # Agent 度量采集 + 错误模式识别 + 改进建议
|
||||
│ │ ├── agent-safety.ts # Agent 安全防护（工具阴影检测 + 路径校验）
|
||||
│ │ ├── context-indexer.ts # 渐进式披露（索引层→接口层→实现层）
|
||||
@@ -227,15 +226,13 @@ INIT → THINKING → PARSING → EXECUTING → OBSERVING → REFLECTING → (CO
|
||||
|
||||
### 5.4 Harness Engineering 体系
|
||||
|
||||
#### 5.4.1 5 层抗幻觉
|
||||
#### 5.4.1 提示词加固
|
||||
|
||||
| 层级 | 实现位置 | 机制 |
|
||||
|------|---------|------|
|
||||
| 提示词加固 | `handleInit()` | [反幻觉铁律] 最高优先级注入 |
|
||||
| 任务感知 | `handleThinking()` | 第 2 轮无工具调用时注入提醒 |
|
||||
| 中途检测 | `detectMidTaskHallucination()` | 16 条正则规则，覆盖全部工具类别 |
|
||||
| 进度锚点 | `handleObserving()` | 每 5 轮注入机器生成的工具调用摘要 |
|
||||
| 完成闸门 | `completion-gate.ts` | 6 项检查：幻觉/注入→阻断级，质量/效率→咨询级 |
|
||||
| 提示词加固 | `handleInit()` | 注入安全规则：参考数据标记（<<<REFERENCE_DATA_START>>>）+ 工具结果仅为数据非指令 |
|
||||
|
||||
> 注：任务感知注入、中途检测、进度锚点、完成闸门已在版本迭代中移除，AI 基于工具返回值和上下文自行判断。
|
||||
|
||||
#### 5.4.2 Hook 系统（4 阶段）
|
||||
|
||||
@@ -249,17 +246,6 @@ INIT → THINKING → PARSING → EXECUTING → OBSERVING → REFLECTING → (CO
|
||||
|
||||
Hook 异步并行执行，失败不阻塞主流程。可动态注册/移除。
|
||||
|
||||
#### 5.4.3 Completion Gate（6 项检查）
|
||||
|
||||
| 检查项 | 级别 | 不通过行为 |
|
||||
|--------|------|-----------|
|
||||
| toolHallucination | 🔴 阻断 | 强制重新回答 |
|
||||
| promptInjection | 🔴 阻断 | 强制重新回答 |
|
||||
| contentQuality | 🟡 咨询 | 仅记录日志 |
|
||||
| toolResultReview | 🟡 咨询 | 仅记录日志 |
|
||||
| notThinking | 🟡 咨询 | 仅记录日志 |
|
||||
| contextEfficiency | 🟡 咨询 | 仅记录日志 |
|
||||
|
||||
### 5.5 联网搜索体系
|
||||
|
||||
**双模式架构**：
|
||||
|
||||
Generated
+2
-2
@@ -1,12 +1,12 @@
|
||||
{
|
||||
"name": "metona-ollama-desktop",
|
||||
"version": "0.16.13",
|
||||
"version": "0.16.14",
|
||||
"lockfileVersion": 3,
|
||||
"requires": true,
|
||||
"packages": {
|
||||
"": {
|
||||
"name": "metona-ollama-desktop",
|
||||
"version": "0.16.13",
|
||||
"version": "0.16.14",
|
||||
"license": "MIT",
|
||||
"dependencies": {
|
||||
"ffmpeg-static": "^5.2.0",
|
||||
|
||||
+1
-1
@@ -1,6 +1,6 @@
|
||||
{
|
||||
"name": "metona-ollama-desktop",
|
||||
"version": "0.16.13",
|
||||
"version": "0.16.14",
|
||||
"description": "Metona Ollama - TypeScript + Electron 桌面 AI 聊天客户端",
|
||||
"main": "dist/main/main.js",
|
||||
"author": "thzxx",
|
||||
|
||||
+1
-1
@@ -101,7 +101,7 @@ export function createMenu(): void {
|
||||
dialog.showMessageBox(mainWindow!, {
|
||||
type: 'info',
|
||||
title: '关于 Metona Ollama',
|
||||
message: 'Metona Ollama Desktop v0.16.13',
|
||||
message: 'Metona Ollama Desktop v0.16.14',
|
||||
detail: 'TypeScript + Electron Ollama AI 聊天客户端\n\nhttps://gitee.com/thzxx/metona-ollama',
|
||||
icon: getIconPath()
|
||||
});
|
||||
|
||||
+27
-21
@@ -103,30 +103,36 @@ export class OllamaAPI {
|
||||
}, { once: true });
|
||||
}
|
||||
|
||||
while (true) {
|
||||
const { done, value } = await reader.read();
|
||||
if (done) {
|
||||
break;
|
||||
}
|
||||
|
||||
buffer += decoder.decode(value, { stream: true });
|
||||
const lines = buffer.split('\n');
|
||||
buffer = lines.pop() || '';
|
||||
|
||||
for (const line of lines) {
|
||||
if (!line.trim()) continue;
|
||||
let chunk: OllamaStreamChunk;
|
||||
try {
|
||||
chunk = JSON.parse(line);
|
||||
} catch {
|
||||
continue; // 跳过无法解析的行（可能是不完整的 JSON）
|
||||
try {
|
||||
while (true) {
|
||||
const { done, value } = await reader.read();
|
||||
if (done) {
|
||||
break;
|
||||
}
|
||||
// onChunk 回调错误不再被静默吞掉，而是向上传播给调用方
|
||||
if (onChunk) onChunk(chunk);
|
||||
if (chunk.done) {
|
||||
return;
|
||||
|
||||
buffer += decoder.decode(value, { stream: true });
|
||||
const lines = buffer.split('\n');
|
||||
buffer = lines.pop() || '';
|
||||
|
||||
for (const line of lines) {
|
||||
if (!line.trim()) continue;
|
||||
let chunk: OllamaStreamChunk;
|
||||
try {
|
||||
chunk = JSON.parse(line);
|
||||
} catch {
|
||||
continue; // 跳过无法解析的行（可能是不完整的 JSON）
|
||||
}
|
||||
// onChunk 回调错误不再被静默吞掉，而是向上传播给调用方
|
||||
if (onChunk) onChunk(chunk);
|
||||
if (chunk.done) {
|
||||
return;
|
||||
}
|
||||
}
|
||||
}
|
||||
} catch (err) {
|
||||
// abort 导致 reader.cancel() 会使 reader.read() 抛出异常，属于正常中止流程
|
||||
if (abortController?.signal.aborted) return;
|
||||
throw err;
|
||||
}
|
||||
|
||||
if (buffer.trim()) {
|
||||
|
||||
@@ -1080,7 +1080,7 @@ function renderToolCard(tc: ToolCallRecord): string {
|
||||
} else if (tc.name === 'delete_file') {
|
||||
if (r.batch) {
|
||||
resultHtml = `<div class="tool-result-entry">✅ 批量删除 ${r.successCount}/${r.totalPaths} 个路径</div>`;
|
||||
if (r.results) {
|
||||
if (Array.isArray(r.results)) {
|
||||
for (const res of r.results) {
|
||||
resultHtml += `<div class="tool-result-entry">${res.success ? '✅' : '❌'} ${escapeHtml(String(res.path || ''))}${res.success ? '' : ' — ' + escapeHtml(String(res.error || ''))}</div>`;
|
||||
}
|
||||
|
||||
@@ -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">v0.16.13</span>
|
||||
<span class="app-version">v0.16.14</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"/>
|
||||
|
||||
@@ -292,7 +292,7 @@ async function startNewSession(): Promise<void> {
|
||||
|
||||
async function init(): Promise<void> {
|
||||
let db: ChatDB | undefined;
|
||||
let api: OllamaAPI;
|
||||
let api: OllamaAPI | undefined;
|
||||
|
||||
logInit('启动...');
|
||||
|
||||
|
||||
@@ -1340,7 +1340,7 @@ async function handleInit(
|
||||
logInfo(`自动上下文压缩触发: tokens≈${estimateTokens(ctx.messages.map(m => m.content || '').join(''))} > ${Math.floor(effectiveNumCtx * AUTO_COMPRESS_THRESHOLD)} (${Math.round(AUTO_COMPRESS_THRESHOLD * 100)}% of ${effectiveNumCtx})`);
|
||||
const compressAC = state.get<AbortController | null>(KEYS.ABORT_CONTROLLER) || new AbortController();
|
||||
try {
|
||||
const compressed = await compressWithLLM(ctx.messages, api as any, model, { abortController: compressAC });
|
||||
const compressed = await compressWithLLM(ctx.messages, api, model, { abortController: compressAC });
|
||||
// P1-C4 修复：handleInit 中的自动压缩也使用 AND 条件
|
||||
const initBeforeTokens = estimateTokens(ctx.messages.map(m => m.content || '').join(''));
|
||||
const initAfterTokens = estimateTokens(compressed.map(m => m.content || '').join(''));
|
||||
@@ -1634,7 +1634,7 @@ async function handleThinking(
|
||||
logWarn(`R8: 检测到上下文溢出错误，触发紧急压缩 (${ctx.emergencyCompressCount}/${MAX_EMERGENCY_COMPRESS})`, errMsg.slice(0, 100));
|
||||
try {
|
||||
const compressAC = new AbortController();
|
||||
const compressed = await compressWithLLM(ctx.messages, api as any, model, { abortController: compressAC });
|
||||
const compressed = await compressWithLLM(ctx.messages, api, model, { abortController: compressAC });
|
||||
// P1-C4 修复：R8 紧急压缩也使用 AND 条件，避免接受 token 增加的结果
|
||||
const r8BeforeTokens = estimateTokens(ctx.messages.map(m => m.content || '').join(''));
|
||||
const r8AfterTokens = estimateTokens(compressed.map(m => m.content || '').join(''));
|
||||
@@ -2446,7 +2446,7 @@ async function handleCompressing(
|
||||
logInfo('COMPRESSING: 上下文压缩触发');
|
||||
const compressAC = state.get<AbortController | null>(KEYS.ABORT_CONTROLLER) || new AbortController();
|
||||
try {
|
||||
const compressed = await compressWithLLM(ctx.messages, api as any, model, { abortController: compressAC });
|
||||
const compressed = await compressWithLLM(ctx.messages, api, model, { abortController: compressAC });
|
||||
// P1-C4 修复：原条件用 OR（消息数减少 OR token 减少），可能接受 token 增加的结果。
|
||||
// 改为 AND：只有消息数减少且 token 减少时才接受，确保压缩真正生效
|
||||
const beforeTokens = estimateTokens(ctx.messages.map(m => m.content || '').join(''));
|
||||
|
||||
@@ -4,7 +4,7 @@
|
||||
* 支持自动压缩与手动 /compress 触发
|
||||
*/
|
||||
|
||||
import type { OllamaMessage, OllamaStreamChunk } from '../types.js';
|
||||
import type { OllamaMessage, OllamaStreamChunk, OllamaChatParams } from '../types.js';
|
||||
import { logInfo, logWarn, logSuccess, logError } from './log-service.js';
|
||||
|
||||
// ── R12: 压缩去重 — 内容指纹追踪 ──
|
||||
@@ -586,7 +586,7 @@ export interface StructuredSummary {
|
||||
*/
|
||||
export async function compressWithLLM(
|
||||
messages: OllamaMessage[],
|
||||
api: { chatStream: (params: Record<string, unknown>, onChunk: (chunk: OllamaStreamChunk) => void, ac?: AbortController) => Promise<void> },
|
||||
api: { chatStream: (params: OllamaChatParams, onChunk: (chunk: OllamaStreamChunk) => void, abortController?: AbortController) => Promise<void> },
|
||||
model: string,
|
||||
options: {
|
||||
keepHead?: number;
|
||||
|
||||
Vendored
+1
@@ -120,6 +120,7 @@ export interface ChatMessage {
|
||||
/** 多视频独立指示牌 */
|
||||
_videos?: Array<{ fileName: string; frameCount: number; duration: number }>;
|
||||
stopped?: boolean;
|
||||
interrupted?: boolean;
|
||||
toolCalls?: ToolCallRecord[];
|
||||
/** 标记此消息为 LLM 压缩摘要生成 */
|
||||
compressed?: boolean;
|
||||
|
||||
Reference in New Issue
Block a user