10次聊天

工程开发 社区

技能解读

当前原文的自动摘译,尚未人工审校。

架构(v0.9.0+)

10x-chat 使用 HTTP 浏览器守护进程 - 持久性本地 Node.js 服务器包装 Playwright:一个守护进程,多个会话:并行 CLI 运行通过 HTTP RPC 共享单个 Chrome 实例 自动启动/停止:守护进程在首次使用时启动,空闲 30 分钟后关闭 崩溃恢复:如果守护进程死亡,下一个 CLI 运行将自动重新启动 无僵尸 Chrome:正确的选项卡引用计数和正常关闭

安装

无需安装。始终使用@latest:使用npx(而不是bunx——并行符号链接冲突)。

何时使用

陷入错误:向另一个模型寻求新的视角 代码审查:将 PR diff 发送到 GPT / Claude / Gemini 进行交叉审查 交叉验证:比较多个模型的答案 知识差距:利用具有不同训练数据的模型 图像生成:通过 ChatGPT 的 DALL-E 或通过 Gemini 的 Imagen

供应商

提供商 · 聊天 · 图像 · 研究 · 视频 · 模型 · 注释 chatgpt · ✅ · ✅ (DALL-E) · ✅ · ❌ · — · 默认运行(反机器人) Gemini · ✅ · ✅ (Imagen) · ✅ · ❌ · 快速、思考(默认)、深度思考(超级工具)、Pro · --model 切换模式 claude · ✅ · ❌ · ❌ · ❌ · — · 默认运行为 grok · ✅ · ✅ · ❌ · ❌ · — · UI 经常改变,使用 @latest

并行会议 (v0.9.0+)

HTTP 守护进程使并行运行稳定。所有提供商共享一台 Chrome:

流狐档案 运行与权限由流狐静态检查推断,不代表作者声明或实际测试结果。
领域
工程开发
作者声明的智能体
未找到明确声明;不据此推断已兼容或已测试
静态检查
88 / 100 · 启发式扫描,不代表运行安全
运行环境证据
未执行;当前没有隔离运行测试证据
作者 / 版本 / 许可证
@MikeChongCan · 未声明许可证
上下文用量评级
低消耗
接入复杂程度
即装即用
是否需要外部接口密钥
不需要
兼容的系统
macOS · Linux · Windows
运行环境要求
Node.js
检测到的文件与系统行为
  • 只读
  • 允许写入 / 修改
网络行为
仅限本地
安装命令数
无(仅作为资料)

需要注意: 未声明工具权限范围;实际权限由运行环境决定。

作者说明文件(原文) SKILL.md

# 10x-chat — AI Agent Skill

Use 10x-chat to send prompts to web-based AI agents via automated browser sessions. Supports chat, image generation, deep research, and video generation. Sessions use a shared persisted profile by default.

## Architecture (v0.9.0+)

10x-chat uses an **HTTP browser daemon** — a persistent local Node.js server wrapping Playwright:

- **One daemon, multiple sessions**: parallel CLI runs share a single Chrome instance via HTTP RPC
- **Auto-start/stop**: daemon launches on first use, shuts down after 30 min idle
- **Crash recovery**: if daemon dies, next CLI run auto-restarts it
- **No zombie Chrome**: proper tab ref-counting and graceful shutdown
- **State file**: `~/.10x-chat/browser-daemon.json` (port, PID, bearer token, chmod 600)

## Installation

No install needed. Always use `@latest`:

```bash
npx 10x-chat@latest --version
```

Use `npx` (not `bunx` — symlink conflicts in parallel).

## When to use

- **Stuck on a bug**: ask another model for a fresh perspective
- **Code review**: send PR diff to GPT / Claude / Gemini for cross-review
- **Cross-validation**: compare answers from multiple models
- **Knowledge gaps**: leverage a model with different training data
- **Image generation**: DALL-E via ChatGPT or Imagen via Gemini
- **Deep research**: long-form analysis via Perplexity, ChatGPT, or Gemini
- **Video generation**: text/image-to-video via Google Flow (Veo) or Dreamina (Seedance)

## Providers

| Provider | Chat | Image | Research | Video | Models | Notes |
|----------|------|-------|----------|-------|--------|-------|
| chatgpt | ✅ | ✅ (DALL-E) | ✅ | ❌ | — | Runs headed by default (anti-bot) |
| gemini | ✅ | ✅ (Imagen) | ✅ | ❌ | Fast, **Thinking** (default), Deep Think (Ultra tool), Pro | `--model` switches mode |
| claude | ✅ | ❌ | ❌ | ❌ | — | Runs headed by default |
| grok | ✅ | ✅ | ❌ | ❌ | — | UI changes often, use `@latest` |
| perplexity | ✅ | ❌ | ✅ | ❌ | — | Best for research with citations |
| notebooklm | ✅ | ❌ | ❌ | ❌ | — | Add sources first, then chat |
| flow | ❌ | ❌ | ❌ | ✅ (Veo) | Veo 3.1 Fast/Quality, Veo 2 Fast/Quality | Google login (shared with Gemini) |
| dreamina | ❌ | ❌ | ❌ | ✅ (Seedance) | Seedance 2.0 Fast (default)/2.0; 1.x often plan-locked | `login dreamina` (CapCut); text + image-to-video |

## Commands

```bash
# Login (one-time per provider — opens browser for auth)
npx 10x-chat@latest login chatgpt
npx 10x-chat@latest login gemini
npx 10x-chat@latest login claude
npx 10x-chat@latest login grok
npx 10x-chat@latest login perplexity
npx 10x-chat@latest login notebooklm

# Chat
npx 10x-chat@latest chat -p "Review this code for bugs" --provider chatgpt --file "src/**/*.ts"
npx 10x-chat@latest chat --provider gemini --file "path/to/prompt.md" -p "Complete this task"
npx 10x-chat@latest chat --provider gemini --model Pro -p "Solve this math problem"
npx 10x-chat@latest chat --provider gemini --model "Deep Think" -p "Solve this hard problem"

# Image generation
npx 10x-chat@latest image -p "A fox astronaut in space" --provider chatgpt
npx 10x-chat@latest image -p "Watercolor landscape" --provider gemini --save-dir ./images

# Deep research (long-form, 5-10 min)
npx 10x-chat@latest research -p "Latest breakthroughs in quantum computing" --provider perplexity
npx 10x-chat@latest research -p "Hard technical research" --provider gemini --model "Deep Think"
npx 10x-chat@latest research -p "Market analysis of EVs" --provider chatgpt --timeout 600000

# Video generation (Flow / Veo default, or Dreamina / Seedance)
npx 10x-chat@latest video -p "Drone shot over snowy peaks at sunrise" --provider flow
npx 10x-chat@latest video -p "Neon street, rain" --provider flow --model "Veo 3.1 - Quality" --orientation portrait
npx 10x-chat@latest login dreamina   # one-time CapCut login for Dreamina
npx 10x-chat@latest video -p "A paper boat in a rain gutter, macro" --provider dreamina --aspect 9:16 --duration 4
npx 10x-chat@latest video -p "The glowing orb floats up" --provider dreamina --image ref.png --ref-mode omni

# Dry run / clipboard
npx 10x-chat@latest chat --dry-run -p "Debug this error" --file src/
npx 10x-chat@latest chat --copy -p "Explain this" --file "src/**"

# Provider chat history + local session management
npx 10x-chat@latest history --provider gemini
npx 10x-chat@latest history --provider all --limit 10
npx 10x-chat@latest status
npx 10x-chat@latest session <id> --render

# NotebookLM
npx 10x-chat@latest notebooklm list
npx 10x-chat@latest notebooklm create "My Research"
npx 10x-chat@latest notebooklm add-url <id> https://...
npx 10x-chat@latest notebooklm add-file <id> ./paper.pdf
npx 10x-chat@latest notebooklm sources <id>
npx 10x-chat@latest notebooklm summarize <id>

# Install bundled skill to coding agent
npx 10x-chat@latest skill install
```

## Parallel sessions (v0.9.0+)

HTTP daemon makes parallel runs stable. All providers share one Chrome:

```bash
# Login all once
npx 10x-chat@latest login gemini
npx 10x-chat@latest login claude
npx 10x-chat@latest login chatgpt

# Run concurrently — each opens a tab in the shared daemon
npx 10x-chat@latest chat --provider gemini -p "Your prompt" --file context.md &
npx 10x-chat@latest chat --provider claude -p "Your prompt" --file context.md &
npx 10x-chat@latest chat --provider chatgpt -p "Your prompt" --file context.md &
wait
```

## Browser mode

- **Headless (default for gemini, grok, perplexity)**: no visible window
- **Headed (default for chatgpt, claude)**: visible Chrome window (anti-bot protection)
- **Force headed**: `--headed` flag on any provider

The daemon stores the headless/headed mode in its state file. If you switch modes, the daemon restarts automatically.

## Profile modes

**Shared (default):** One browser profile, all providers share cookies. Login once per Google account covers Gemini + NotebookLM.

**Isolated:** Separate profile per provider (backward compat): `--isolated-profile`

```bash
# Migrate from isolated to shared
npx 10x-chat@latest migrate
```

## Daemon management

```bash
# Check daemon state
cat ~/.10x-chat/browser-daemon.json

# Force stop daemon
# (CLI calls stopDaemon() on Ctrl+C automatically)
kill $(cat ~/.10x-chat/browser-daemon.json | python3 -c "import sys,json; print(json.load(sys.stdin)['pid'])")
```

## Tips

- **Always use `@latest`**: ensures newest fixes
- **Login first**: `login <provider>` once per provider. Sessions persist in `~/.10x-chat/profiles/`
- **Use `--headed`** if a provider is flaky (Grok especially)
- **Keep file sets small**: fewer files + focused prompt = better answers
- **Research needs longer timeouts**: `--timeout 600000` for 10-min research jobs
- **Image gen can take 1-2 min**: use `--timeout 120000` when needed
- **Video gen can take 1-5 min**: Dreamina queues generations; keep the default 10-min timeout. For image-to-video, pass `--image` (Dreamina) or `--start-frame`/`--end-frame` with `--mode frames` (Flow)
- **Dreamina models are plan-gated**: `Seedance 2.0 Fast` (cheapest) and `2.0` are generally available; the CLI errors clearly if a requested model is locked
- **Use `--dry-run`** to preview what will be sent

## Known issues

- **Grok**: UI changes frequently. Always use `@latest`. Use `--headed` for best reliability
- **ChatGPT/Grok sessions expire quickly**: login again if you get "Not logged in" errors
- **Some provider UIs are flaky under automation**: retry with `--headed` before assuming a hard failure

## Safety

- Never include credentials, API keys, or tokens in the bundled files
- The tool opens a real browser with real login state. Treat it like your own browser session

讨论

基于 GitHub Discussions。登录 GitHub 即可参与讨论、点赞、订阅更新。