tts-winmsg-free

Windows本地TTS朗读技能。通过WM_COPYDATA消息把AI回复实时发送到迷你窗口朗读，支持中英文标点分割、队列顺序播放、语速音量滑块调节，零token消耗，完全本地运行。

Security Scan

This skill is internally consistent with a local Windows TTS helper: it provides local Python scripts to host a small window and send text to it via WM_COPYDATA; it does not request credentials or network access and the code matches the SKILL.md instructions.

Purpose & Capability

Name/description (local Windows TTS via WM_COPYDATA) match the included scripts and config files. The files implement a small always-on-top Tk window, save its HWND, and a sender script that reads that HWND and sends text; these are appropriate for the stated purpose.

Instruction Scope

SKILL.md requires the agent to run local scripts on every user-visible reply (python sendmsg_hwnd.py <text>) and to start the wmserver process before use. This is within the scope of a local TTS skill, but it gives the agent the ability to execute a local command on every reply — users should be aware the agent will run these processes and that every visible reply will be sent to the local TTS window (spoken aloud).

Install Mechanism

There is no automated install spec; the README-style instructions ask the user to pip install pywin32 and run the scripts. No external downloads or opaque installers are present in the package.

Credentials

The skill requests no environment variables or credentials. It writes and reads local config files (tts_speed.txt, volume.txt, wmserver_hwnd.txt). A local risk: if the HWND file is tampered with (pointing to another process's window), the sender will forward text to that HWND — WM_COPYDATA can reach other processes, so file integrity matters. No network exfiltration or secret access is present in the code.

Persistence & Privilege

The skill does not request 'always' inclusion, does not modify other skills, and keeps state only in its own config directory. It runs as local processes only when started by the user/agent.

Assessment

This skill appears to do what it says: run a small local TTS window and send your AI's visible replies to it. Before installing/running, consider: (1) it requires Windows, Python, and pywin32; you must start the wmserver script once and the agent will run sendmsg_hwnd.py for each visible reply (so anything you see in the chat will also be spoken aloud). (2) The scripts read/write config/wmserver_hwnd.txt — if that file is maliciously changed to an arbitrary HWND, the sender could forward text to another process; keep the config directory trusted. (3) The code runs PowerShell to speak text locally (no network calls or credential access). If you obtained the skill from an untrusted source, review the scripts yourself or run them in an isolated account before use.

winmsg-tts

Windows 本地 TTS 朗读技能。通过 WM_COPYDATA 消息把 AI 回复实时发送到迷你窗口朗读，支持中英文标点分割、队列顺序播放、语速音量滑块调节，零 token 消耗，完全本地运行。

核心规则（重要！）

开启朗读模式

• 用户说"开启朗读模式"时 → 自动启动 TTS 迷你窗口
• 启动后第一条消息要说点自然的、有上下文的话，不要说通用测试句
• 把 HWND 和启动成功状态告诉用户

朗读中（每次回复必须执行！）

TTS 模式下，AI 的每一条用户可见回复，都必须同步调用 sendmsg_hwnd.py 发送到小窗口朗读。不得遗漏。不得只发文字不转发音频。

流程：

1. 收到用户消息
2. 生成回复文字
3. 调用 sendmsg_hwnd.py：python sendmsg_hwnd.py <回复内容>
4. 把同一回复文字也发给用户（webchat）

关闭朗读模式

• 用户说"关闭朗读模式"时 → 发送精确退出命令：python sendmsg_hwnd.py --quit
• 只关闭自己的 TTS 窗口，不会影响其他进程
• 关闭后告知用户

规则要点

• 只把用户看得见的回复发给 TTS 窗口，不要发内部推理
• 开启时说自然的话，不要每次都说"这是一条测试消息"
• 滑块调节实时生效，无需重启

技术特性

• 自动分句：按中英文标点（。 ； ： ！ ？ …… . , ; : ! ?）分割句子，保留标点到句尾
• 队列播放：多句按顺序排队播放，不重叠，不打断
• label 同步：窗口文字随播放进度同步更新，每播一句更新一句
• 无弹窗：使用 CREATE_NO_WINDOW (0x08000000) + Base64 编码，彻底解决 PowerShell 窗口问题
• 配置持久化：语速/音量保存到配置文件，断电重启后恢复上次设置
• 精确关闭：通过 __QUIT__ 命令关闭自己的窗口，不杀其他进程

文件说明

• scripts/wmserver_20260327.py — TTS 接收端，320×90像素迷你窗口，无边框右下角永远置顶
• scripts/sendmsg_hwnd.py — 发送端，通过 WM_COPYDATA 发送字符串；支持 --quit 精确关闭
• config/tts_speed.txt — 语速配置（-10 到 +10）
• config/volume.txt — 音量配置（0-100）
• config/wmserver_hwnd.txt — 当前窗口 HWND（运行时生成）

使用方法

1. 安装依赖：pip install pywin32
2. 启动 TTS：python scripts/wmserver_20260327.py
3. 发送信息：python scripts/sendmsg_hwnd.py <消息内容>
4. 关闭 TTS：python scripts/sendmsg_hwnd.py --quit

技术原理

• WM_COPYDATA = 0x004A + COPYDATASTRUCT 结构，跨进程传字符串
• PowerShell System.Speech.Synthesis.SpeechSynthesizer TTS 引擎
• Base64 编码命令避免中文和长度限制
• 消息队列顺序播放，不重叠
• tkinter 做 UI，overrideredirect(True) 无边框窗口
• CREATE_NO_WINDOW 禁止 PowerShell 弹窗
• __QUIT__ 命令通过窗口类名精确关闭窗口，不影响其他进程
• 配置持久化到文本文件，断电重启后恢复上次设置

UI 布局（320×90）

[白色文字显示区域（当前播放的句子）]
[慢]━━━●━━━━━━━━[快]   ← 语速滑块
[小]━━━●━━━━━━━━[大]   ← 音量滑块

• 背景：深灰色 #2d2d2d
• 文字：白色 #FFFFFF，微软雅黑 10pt 粗体
• 滑块轨道：灰色 #555555
• 滑块把手：白色 oval

系统要求

• Windows 系统
• Python 3.8+
• pywin32