skyloom 1.11.0 → 1.13.0

package/.github/workflows/ci.yml

@@ -1,36 +1,36 @@
-name: CI
-
-on:
-  push:
-    branches: [ main, master ]
-  pull_request:
-    branches: [ main, master ]
-
-jobs:
-  build-and-test:
-    runs-on: ubuntu-latest
-
-    strategy:
-      matrix:
-        node-version: [18, 20, 22]
-
-    steps:
-    - uses: actions/checkout@v4
-
-    - name: Use Node.js ${{ matrix.node-version }}
-      uses: actions/setup-node@v4
-      with:
-        node-version: ${{ matrix.node-version }}
-        cache: 'npm'
-
-    - name: Install dependencies
-      run: npm ci
-
-    - name: TypeScript compile check
-      run: npx tsc --noEmit
-
-    - name: Run tests
-      run: npx vitest run
-
-    - name: Build
-      run: npm run build
+name: CI
+
+on:
+  push:
+    branches: [ main, master ]
+  pull_request:
+    branches: [ main, master ]
+
+jobs:
+  build-and-test:
+    runs-on: ubuntu-latest
+
+    strategy:
+      matrix:
+        node-version: [18, 20, 22]
+
+    steps:
+    - uses: actions/checkout@v4
+
+    - name: Use Node.js ${{ matrix.node-version }}
+      uses: actions/setup-node@v4
+      with:
+        node-version: ${{ matrix.node-version }}
+        cache: 'npm'
+
+    - name: Install dependencies
+      run: npm ci
+
+    - name: TypeScript compile check
+      run: npx tsc --noEmit
+
+    - name: Run tests
+      run: npx vitest run
+
+    - name: Build
+      run: npm run build

package/README.md

@@ -1,67 +1,163 @@
 # 天空织机 · Skyloom
 
-**6 个天气主题 AI Agent** — 雾·雨·霜·雪·露·晴
+<div align="center">
 
-## 一键安装
+**≋ 雾 · ⸽ 雨 · ✱ 霜 · ❉ 雪 · ∘ 露 · ☼ 晴**
+
+*六位 Agent，一支团队 —— 一座水墨气象台。*
+
+[![CI](https://github.com/susurrune/skyloom-ts/actions/workflows/ci.yml/badge.svg)](https://github.com/susurrune/skyloom-ts/actions)
+[![npm](https://img.shields.io/npm/v/skyloom?color=c0392b)](https://www.npmjs.com/package/skyloom)
+[![License: MIT](https://img.shields.io/badge/license-MIT-3a7a6e.svg)](LICENSE)
+
+</div>
+
+---
+
+**Skyloom** 是一个本地优先的多智能体终端框架。六个 Agent 各司其职，通过事件总线、三层记忆、DAG 编排引擎协作完成完整工作流。不是又一个 LLM 聊天客户端，而是一个**分工明确的 AI 团队**——而且，它好看。
+
+每个 Agent 是一种天气、一味矿物颜料、一句古诗。这套贯穿 CLI、TUI 与 Web 的「水墨气象台」意象，是 Skyloom 的辨识度所在。设计理念见 **[美学设计系统](docs/AESTHETIC_DESIGN.md)**。
+
+## 安装
 
 ```bash
-# 方式一：项目内安装（自动全局注册）
-git clone https://github.com/susurrune/skyloom-ts.git
-cd skyloom-ts
-npm run setup               # ← 一条命令：安装 → 编译 → 注册全局 sky 命令
-
-# 方式二：全局安装（像 Claude Code 一样）
-npm install -g skyloom       # 发布后
-# 或本地：
-npm install -g ./             # 装完后终端输入 sky 即可使用
-
-# 方式三：直接运行（免安装）
-npx skyloom
+npm install -g skyloom
+sky
 ```
 
-装完后在任意终端输入 `sky` 就能用。
+首次启动自动进入设置向导 — 选 provider → 输 API Key → 选模型，全部交互式完成。Key 保存在 `~/.skyloom/config.yaml`，更新不丢。随时 `/setup` 切换。
+
+### 支持的 Provider（9 家）
+
+| Provider | 可调模型 |
+|----------|------|
+| **DeepSeek** | deepseek-chat, deepseek-reasoner, deepseek-v4-flash, deepseek-v4-pro |
+| **OpenAI** | gpt-4.1, gpt-4.1-mini, gpt-4o, gpt-4o-mini, o4-mini |
+| **Anthropic** | claude-sonnet-4-6, claude-opus-4-7, claude-haiku-4-5 |
+| **Google** | gemini-2.5-pro, gemini-2.5-flash |
+| **Groq** | llama-4-scout, llama-3.3-70b, mixtral-8x7b |
+| **OpenRouter** | gpt-4.1, claude-sonnet-4-6, gemini-2.5-flash, llama-4-maverick |
+| **Mistral** | mistral-large, mistral-small |
+| **xAI** | grok-4 |
+| **Ollama** | llama3, qwen2.5, deepseek-r1（本地） |
+
+> 模型目录是单一事实源，含上下文窗口与成本数据，参见 [`config/models.yaml`](config/models.yaml)。
 
 ## 命令
 
-| 命令 | 用途 |
+```bash
+sky                  # 全屏 TUI 交互对话
+sky fog              # 直接和雾对话
+sky task "写一个CLI"  # 多 Agent 编排
+sky web              # Web UI → http://localhost:3000
+sky apikey set <p>   # 保存 API Key
+sky config           # 查看配置
+sky mcp              # MCP Server
+sky init             # 初始化
+```
+
+## TUI
+
+输入 `/` 弹出命令面板，↑↓ 选择，字母过滤，Enter 确认：
+
+```
+┌── 天空织机 v1.13 ──────────────────┐
+├──────────┬─────────────────────────┤
+│ ≋ 雾 Fog│                         │
+│ ⸽ 雨 ▸  │  用户消息                │
+│ ✱ 霜    │                         │
+│ ❉ 雪    │  助手回复                │
+│ ∘ 露    │                         │
+│ ☼ 晴    │                         │
+├──────────┴─────────────────────────┤
+│ ┌── commands ────────────────────┐ │
+│ ▶ ≋ /fog    雾 Fog · 松烟墨      │ │
+│     /setup  配置向导              │ │
+│     /cost   费用统计              │ │
+│ └────────────────────────────────┘ │
+│ > /                                 │
+└─────────────────────────────────────┘
+```
+
+## Slash 命令
+
+| 类别 | 命令 |
 |------|------|
-| `sky chat` | 交互式对话（默认雾） |
-| `sky chat fog` | 指定 Agent 对话 |
-| `sky web` | 启动水墨气象台 Web UI → http://localhost:3000 |
-| `sky task "写一个 CLI 工具"` | 多 Agent 编排协作 |
-| `sky config` | 查看配置 |
-| `sky mcp` | MCP Server（Claude Desktop 可连接） |
-| `sky init` | 初始化配置 |
-
-## 六灵 · Agents
-
-|  | 名称 | 矿物色 | 职责 |
-|--|------|--------|------|
-| ≋ | **雾** Fog | 松烟墨 | 探索洞察 · 研究 |
-| ⸽ | **雨** Rain | 石青 | 创造产出 · 代码 |
-| ✱ | **霜** Frost | 石绿 | 精炼品质 · 审查 |
-| ❉ | **雪** Snow | 铅白 | 架构规划 · 编排 |
-| ∘ | **露** Dew | 赭石 | 可靠守护 · 运维 |
-| ☼ | **晴** Fair | 朱砂 | 情感陪伴 · 温暖 |
+| **Agent** | `/fog /rain /frost /snow /dew /fair` — 切换 |
+| **配置** | `/setup /apikey /model` — 安装向导、设置 Key、模型 |
+| **信息** | `/status /cost /memory /sessions /workspace /version` |
+| **操作** | `/compact /retry /clear /task <goal>` |
+| **退出** | `/quit /exit` |
 
-## 开发
+## 六灵
+
+| Agent | 矿物色 | 职责 | 技能 |
+|-------|--------|------|------|
+| ≋ **雾** Fog | 松烟墨 `#4a4a44` | 探索洞察 | web_research, code_analysis |
+| ⸽ **雨** Rain | 石青 `#2a5c8a` | 创造产出 | code_generator, content_writer |
+| ✱ **霜** Frost | 石绿 `#3a7a6e` | 精炼品质 | code_reviewer, security_auditor |
+| ❉ **雪** Snow | 铅白 `#8a8a82` | 架构规划 | task_planner, arch_designer |
+| ∘ **露** Dew | 赭石 `#8b6914` | 可靠守护 | sys_operator, ci_cd_manager |
+| ☼ **晴** Fair | 朱砂 `#b3342d` | 情感陪伴 | emotional_companion, self_evolve |
+
+## 核心能力
+
+| 模块 | 说明 |
+|------|------|
+| **真流式** | CLI 逐字渲染、Web SSE 推流；推理/正文/工具调用分层呈现 |
+| **智能路由** | 输入自动分 direct/single/orchestrate 三级，< 1ms |
+| **Pipeline 模板** | 9 种预定义工作流，命中后跳过 LLM 拆解 |
+| **三层记忆** | Short-term(SQLite) + Working(内存) + Long-term(持久化) |
+| **技能系统** | 17 个内置 SKILL.md，运行时动态激活 |
+| **安全体系** | 5 级危险等级、红线拦截、沙箱隔离、审计日志 |
+| **自进化** | 失败模式分析、Prompt 自动优化、经验库去重 |
+| **向量搜索** | TF-IDF + Cosine，零依赖语义检索 |
+| **知识图谱** | 实体-关系三重存储，自动从对话提取 |
+| **输出过滤** | API Key/密码/私钥/邮箱/内网IP 自动脱敏 |
+| **电脑操作** | 10 个跨平台工具（应用/诊断/进程/软件/服务） |
+| **MCP** | 双向桥：Client 连接外部 + Server 暴露给 Claude Desktop |
+
+## Web UI · 水墨气象台
 
 ```bash
-npm test          # 87 测试
-npm run build     # 编译
-npm run dev       # watch 模式
-npm install -g ./ # 本地全局安装
+sky web
+# 打开 http://localhost:3000
 ```
 
+宣纸质感、六矿物颜料、按 Agent 切换的气象粒子（雾飘/雨落/霜结/雪降/露凝/晴升）与印章汉字。`⌘1-6` 唤灵切换。回复经 SSE **真流式**推送，工具调用呈现为"气象事件"。
+
 ## 架构
 
 ```
 src/
-├── core/         核心框架（agent, llm, memory, tool, bus...）
-├── agents/       6 个 Agent 实现
-├── tools/        内置工具（文件、Shell、HTTP、Git、系统操作）
-├── cli/          命令行界面
-├── web/          Web 服务器 + 水墨 UI
-├── skills/       技能加载器
-└── plugins/      插件加载器
+├── cli/        main.ts (命令注册+TUI循环), tui.ts (全屏渲染+弹窗), mode.ts
+├── core/       agent, factory, llm, memory, tool, bus, security, learn,
+│              evolve, sandbox, vector, graph, filter, estimate, arbitrate, ...
+├── agents/     6 Agent (fog/rain/frost/snow/dew/fair)
+├── tools/      builtin, computer, delegate
+├── web/        server (水墨气象台 HTTP+UI), tts
+├── skills/     技能加载器
+├── plugins/    插件加载器
+config/
+├── skills/     17 个内置 SKILL.md
+├── default.yaml  providers.yaml  models.yaml
+tests/           13 套件 · 129 Vitest 用例
+docs/            OPTIMIZATION_PLAN.md  AESTHETIC_DESIGN.md
 ```
+
+## 开发
+
+```bash
+npm test          # vitest
+npm run build     # tsc
+npm run dev       # watch
+npm run type-check
+```
+
+## 路线图
+
+Skyloom 正朝「顶级开源 Agent 框架」演进，对标 [opencode](https://github.com/sst/opencode) 的架构与 Claude Code 的交互范式。完整规划见 **[优化计划](docs/OPTIMIZATION_PLAN.md)**：统一模型目录 → 真流式 → `agent.ts` 分层 → Session 与自动上下文压缩 → 工具/插件健壮性 → 测试门禁 → 美学工程化。
+
+---
+
+**[MIT License](LICENSE)** · **v1.13.0** · 全功能迁移自 [Python 原版](https://github.com/susurrune/skyloom)

package/config/default.yaml

@@ -1,47 +1,43 @@
-# Skyloom default configuration
-llm:
-  default_model: gpt-4o
-  language: zh
-  max_retries: 2
-  temperature: 0.7
-  max_tokens: 4096
-
-agents:
-  fog:
-    model: gpt-4o
-    temperature: 0.7
-  rain:
-    model: gpt-4o
-    temperature: 0.7
-  frost:
-    model: gpt-4o
-    temperature: 0.3
-  snow:
-    model: gpt-4o
-    temperature: 0.5
-  dew:
-    model: gpt-4o
-    temperature: 0.3
-  fair:
-    model: gpt-4o
-    temperature: 0.9
-
-memory:
-  db_path: ~/.skyloom/memory.db
-  short_term_limit: 100
-  max_persisted_messages: 2000
-
-workspace:
-  path: auto
-
-cli:
-  default_agent: fog
-  approval_mode: interactive
-
-plugins:
-  enabled: true
-  directories:
-    - ~/.skyloom/plugins
-
-mcp:
-  servers: []
+# Skyloom default configuration
+llm:
+  default_model: gpt-4o
+  language: zh
+  max_retries: 2
+  temperature: 0.7
+  max_tokens: 4096
+
+# Per-agent models inherit `llm.default_model` (or the user's wizard-chosen
+# top-level default_model) unless an agent sets its own `model:` here.
+agents:
+  fog:
+    temperature: 0.7
+  rain:
+    temperature: 0.7
+  frost:
+    temperature: 0.3
+  snow:
+    temperature: 0.5
+  dew:
+    temperature: 0.3
+  fair:
+    temperature: 0.9
+
+memory:
+  db_path: ~/.skyloom/memory.db
+  short_term_limit: 100
+  max_persisted_messages: 2000
+
+workspace:
+  path: auto
+
+cli:
+  default_agent: fog
+  approval_mode: interactive
+
+plugins:
+  enabled: true
+  directories:
+    - ~/.skyloom/plugins
+
+mcp:
+  servers: []

package/config/models.yaml

@@ -0,0 +1,155 @@
+# Model Catalog — latest models per provider
+# Every model listed here is directly callable via the LLM client.
+
+openai:
+  - name: gpt-4.1
+    context: 1048576
+    cost_in: 2.00
+    cost_out: 8.00
+    desc: "旗舰，百万token上下文"
+  - name: gpt-4.1-mini
+    context: 1048576
+    cost_in: 0.40
+    cost_out: 1.60
+    desc: "轻量，同款长上下文"
+  - name: gpt-4o
+    context: 131072
+    cost_in: 2.50
+    cost_out: 10.00
+    desc: "多模态全能"
+  - name: gpt-4o-mini
+    context: 131072
+    cost_in: 0.15
+    cost_out: 0.60
+    desc: "小模型高性价比"
+  - name: o4-mini
+    context: 204800
+    cost_in: 1.10
+    cost_out: 4.40
+    desc: "推理模型轻量版"
+
+anthropic:
+  - name: claude-sonnet-4-6
+    context: 204800
+    cost_in: 3.00
+    cost_out: 15.00
+    desc: "编程+推理首选"
+  - name: claude-opus-4-7
+    context: 204800
+    cost_in: 5.00
+    cost_out: 25.00
+    desc: "最强旗舰"
+  - name: claude-haiku-4-5
+    context: 204800
+    cost_in: 0.80
+    cost_out: 4.00
+    desc: "速度最快"
+
+deepseek:
+  - name: deepseek-chat
+    context: 131072
+    cost_in: 0.27
+    cost_out: 1.10
+    desc: "通用对话"
+  - name: deepseek-reasoner
+    context: 65536
+    cost_in: 0.55
+    cost_out: 2.19
+    desc: "深度推理 (alias → v4-flash)"
+  - name: deepseek-v4-flash
+    context: 131072
+    cost_in: 0.14
+    cost_out: 0.28
+    desc: "极速低成本 (deepseek-chat 实际后端)"
+  - name: deepseek-v4-pro
+    context: 131072
+    cost_in: 1.74
+    cost_out: 3.48
+    desc: "旗舰版"
+
+google:
+  - name: gemini-2.5-pro
+    context: 1048576
+    cost_in: 1.25
+    cost_out: 10.00
+    desc: "旗舰推理"
+  - name: gemini-2.5-flash
+    context: 1048576
+    cost_in: 0.30
+    cost_out: 2.50
+    desc: "快速长上下文"
+
+groq:
+  - name: llama-4-scout
+    context: 131072
+    cost_in: 0
+    cost_out: 0
+    desc: "Meta最新，免费额度"
+  - name: llama-3.3-70b
+    context: 131072
+    cost_in: 0
+    cost_out: 0
+    desc: "Llama旗舰"
+  - name: mixtral-8x7b
+    context: 32768
+    cost_in: 0
+    cost_out: 0
+    desc: "MoE混合专家"
+
+mistral:
+  - name: mistral-large
+    context: 131072
+    cost_in: 2.00
+    cost_out: 6.00
+    desc: "旗舰通用"
+  - name: mistral-small
+    context: 32768
+    cost_in: 0.20
+    cost_out: 0.60
+    desc: "轻量高效"
+
+openrouter:
+  - name: openai/gpt-4.1
+    context: 1048576
+    cost_in: 2.00
+    cost_out: 8.00
+    desc: "通过OpenRouter"
+  - name: anthropic/claude-sonnet-4-6
+    context: 204800
+    cost_in: 3.00
+    cost_out: 15.00
+    desc: "通过OpenRouter"
+  - name: google/gemini-2.5-flash
+    context: 1048576
+    cost_in: 0.30
+    cost_out: 2.50
+    desc: "通过OpenRouter"
+  - name: meta-llama/llama-4-maverick
+    context: 131072
+    cost_in: 0.20
+    cost_out: 0.60
+    desc: "Meta旗舰"
+
+xai:
+  - name: grok-4
+    context: 131072
+    cost_in: 3.00
+    cost_out: 15.00
+    desc: "Grok最新"
+
+ollama:
+  - name: llama3
+    context: 8192
+    cost_in: 0
+    cost_out: 0
+    desc: "本地部署"
+  - name: qwen2.5
+    context: 32768
+    cost_in: 0
+    cost_out: 0
+    desc: "通义千问"
+  - name: deepseek-r1
+    context: 32768
+    cost_in: 0
+    cost_out: 0
+    desc: "本地推理"

package/config/providers.yaml

@@ -1,39 +1,39 @@
-# Provider catalog — API key env vars, base URLs, docs
-openai:
-  env_var: OPENAI_API_KEY
-  base_url: https://api.openai.com/v1
-  docs_url: https://platform.openai.com/api-keys
-
-anthropic:
-  env_var: ANTHROPIC_API_KEY
-  base_url: https://api.anthropic.com/v1
-  docs_url: https://console.anthropic.com/settings/keys
-
-deepseek:
-  env_var: DEEPSEEK_API_KEY
-  base_url: https://api.deepseek.com/v1
-  docs_url: https://platform.deepseek.com/api_keys
-
-groq:
-  env_var: GROQ_API_KEY
-  base_url: https://api.groq.com/openai/v1
-
-mistral:
-  env_var: MISTRAL_API_KEY
-  base_url: https://api.mistral.ai/v1
-
-cohere:
-  env_var: COHERE_API_KEY
-  base_url: https://api.cohere.ai/v1
-
-openrouter:
-  env_var: OPENROUTER_API_KEY
-  base_url: https://openrouter.ai/api/v1
-
-gemini:
-  env_var: GEMINI_API_KEY
-  base_url: https://generativelanguage.googleapis.com/v1beta
-
-ollama:
-  base_url: http://localhost:11434/v1
-  env_var: OLLAMA_HOST
+# Provider catalog — API key env vars, base URLs, docs
+openai:
+  env_var: OPENAI_API_KEY
+  base_url: https://api.openai.com/v1
+  docs_url: https://platform.openai.com/api-keys
+
+anthropic:
+  env_var: ANTHROPIC_API_KEY
+  base_url: https://api.anthropic.com/v1
+  docs_url: https://console.anthropic.com/settings/keys
+
+deepseek:
+  env_var: DEEPSEEK_API_KEY
+  base_url: https://api.deepseek.com/v1
+  docs_url: https://platform.deepseek.com/api_keys
+
+groq:
+  env_var: GROQ_API_KEY
+  base_url: https://api.groq.com/openai/v1
+
+mistral:
+  env_var: MISTRAL_API_KEY
+  base_url: https://api.mistral.ai/v1
+
+cohere:
+  env_var: COHERE_API_KEY
+  base_url: https://api.cohere.ai/v1
+
+openrouter:
+  env_var: OPENROUTER_API_KEY
+  base_url: https://openrouter.ai/api/v1
+
+gemini:
+  env_var: GEMINI_API_KEY
+  base_url: https://generativelanguage.googleapis.com/v1beta
+
+ollama:
+  base_url: http://localhost:11434/v1
+  env_var: OLLAMA_HOST

package/config/skills/api_integrator/SKILL.md

@@ -1,15 +1,15 @@
----
-name: api_integrator
-description: "API integration specialist — REST, GraphQL, OAuth, webhooks"
-triggers:
-  - "api"
-  - "integration"
-  - "rest"
-  - "graphql"
-  - "oauth"
-  - "webhook"
----
-
-## API integration specialist
-
-Integrate external REST/GraphQL APIs. Use http_get and http_post to explore endpoints. Verify responses before writing integration code. Always handle errors and rate limits.
+---
+name: api_integrator
+description: "API integration specialist — REST, GraphQL, OAuth, webhooks"
+triggers:
+  - "api"
+  - "integration"
+  - "rest"
+  - "graphql"
+  - "oauth"
+  - "webhook"
+---
+
+## API integration specialist
+
+Integrate external REST/GraphQL APIs. Use http_get and http_post to explore endpoints. Verify responses before writing integration code. Always handle errors and rate limits.

package/config/skills/arch_designer/SKILL.md

@@ -1,13 +1,13 @@
----
-name: arch_designer
-description: "Software architecture design — patterns, trade-offs, system diagrams"
-triggers:
-  - "architecture"
-  - "design"
-  - "system design"
-  - "patterns"
----
-
-## Software architecture design
-
-Design software architecture with clear trade-offs. Present 2-3 options with pros/cons. Use write_file for architecture decision records.
+---
+name: arch_designer
+description: "Software architecture design — patterns, trade-offs, system diagrams"
+triggers:
+  - "architecture"
+  - "design"
+  - "system design"
+  - "patterns"
+---
+
+## Software architecture design
+
+Design software architecture with clear trade-offs. Present 2-3 options with pros/cons. Use write_file for architecture decision records.