diff --git a/.github/workflows/release.yml b/.github/workflows/release.yml
index d310b33..5d3b979 100644
--- a/.github/workflows/release.yml
+++ b/.github/workflows/release.yml
@@ -14,14 +14,14 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Checkout
-        uses: actions/checkout@v4
+        uses: actions/checkout@v5
         with:
           fetch-depth: 0
 
       - name: Set up Go
-        uses: actions/setup-go@v5
+        uses: actions/setup-go@v6
         with:
-          go-version: '1.19'
+          go-version: '1.26'
 
       - name: Run tests
         run: go test -v ./...
diff --git a/.github/workflows/test.yml b/.github/workflows/test.yml
index 379d900..b2ee6c3 100644
--- a/.github/workflows/test.yml
+++ b/.github/workflows/test.yml
@@ -11,15 +11,15 @@ jobs:
     strategy:
       matrix:
         os: [ubuntu-latest, macos-latest]
-        go-version: ['1.19', '1.20', '1.21']
+        go-version: ['1.26']
     runs-on: ${{ matrix.os }}
 
     steps:
       - name: Checkout code
-        uses: actions/checkout@v4
+        uses: actions/checkout@v5
 
       - name: Set up Go
-        uses: actions/setup-go@v5
+        uses: actions/setup-go@v6
         with:
           go-version: ${{ matrix.go-version }}
 
@@ -39,22 +39,22 @@ jobs:
       - name: Run tests
         run: go test -v -race -coverprofile=coverage.txt -covermode=atomic ./...
 
-      - name: Upload coverage
-        if: matrix.os == 'ubuntu-latest' && matrix.go-version == '1.21'
-        uses: codecov/codecov-action@v3
-        with:
-          file: ./coverage.txt
+      # - name: Upload coverage
+      #   if: matrix.os == 'ubuntu-latest' && matrix.go-version == '1.26'
+      #   uses: codecov/codecov-action@v3
+      #   with:
+      #     file: ./coverage.txt
 
   # lint:
   #   runs-on: ubuntu-latest
   #   steps:
   #     - name: Checkout code
-  #       uses: actions/checkout@v4
+  #       uses: actions/checkout@v5
 
   #     - name: Set up Go
-  #       uses: actions/setup-go@v5
+  #       uses: actions/setup-go@v6
   #       with:
-  #         go-version: '1.21'
+  #         go-version: '1.26'
 
   #     - name: Run golangci-lint
   #       uses: golangci/golangci-lint-action@v3
@@ -85,12 +85,12 @@ jobs:
 
     steps:
       - name: Checkout code
-        uses: actions/checkout@v4
+        uses: actions/checkout@v5
 
       - name: Set up Go
-        uses: actions/setup-go@v5
+        uses: actions/setup-go@v6
         with:
-          go-version: '1.21'
+          go-version: '1.26'
 
       - name: Build
         env:
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 7b453db..b7cc2d3 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -19,7 +19,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Built-in tools: file operations, shell execution, code search
 - Permission confirmation system with risk classification
 - Multi-LLM provider support (Anthropic Claude, OpenAI GPT)
-- Project context awareness (AICODER.md, Git status, dependencies)
+- Project context awareness (.AICODER.md, Git status, dependencies)
 - 11 slash commands: /help, /clear, /history, /undo, /diff, /commit, /cost, /model, /config, /init, /exit
 - Session management with snapshots and undo
 - Token usage tracking and cost estimation
diff --git a/README.md b/README.md
index 812c807..c77bb58 100644
--- a/README.md
+++ b/README.md
@@ -22,7 +22,7 @@
 - **内置工具集**：文件读写编辑、Shell 命令执行、全局代码搜索
 - **权限确认机制**：危险操作前要求用户确认，支持黑名单和白名单
 - **多 LLM 提供商**：Anthropic Claude、OpenAI GPT（兼容任何 OpenAI 格式端点）
-- **项目上下文感知**：自动读取 `AICODER.md`、Git 状态、项目依赖
+- **项目上下文感知**：自动读取 `.AICODER.md`、Git 状态、项目依赖
 - **斜杠命令**：`/diff`、`/undo`、`/commit`、`/cost` 等 11 个内置命令
 - **纯 Go 标准库**：无外部依赖，单二进制，跨平台
 
@@ -113,7 +113,7 @@ $ aicoder
 | `/cost` | 查看 Token 用量和费用估算 |
 | `/model [name]` | 查看或切换 AI 模型 |
 | `/config` | 查看当前配置 |
-| `/init` | 在当前目录生成 AICODER.md 模板 |
+| `/init` | 在当前目录生成 .AICODER.md 模板 |
 | `/exit` | 退出程序 |
 
 ---
@@ -143,9 +143,9 @@ $ aicoder
 
 ---
 
-## AICODER.md 项目配置
+## .AICODER.md 项目配置
 
-在项目根目录创建 `AICODER.md`（或运行 `/init`），AI 会在每次会话开始时自动加载它作为项目级系统提示词：
+在项目根目录创建 `.AICODER.md`（或运行 `/init`），AI 会在每次会话开始时自动加载它作为项目级系统提示词：
 
 ```markdown
 # 项目说明
diff --git a/cmd/root.go b/cmd/root.go
index 19f3e74..72f7176 100644
--- a/cmd/root.go
+++ b/cmd/root.go
@@ -10,13 +10,14 @@ import (
 	"strings"
 	"syscall"
 
-	"github.com/chzyer/readline"
 	"github.com/iminders/aicoder/internal/agent"
 	"github.com/iminders/aicoder/internal/config"
 	"github.com/iminders/aicoder/internal/llm"
 	anthropicprovider "github.com/iminders/aicoder/internal/llm/anthropic"
+	deepseekprovider "github.com/iminders/aicoder/internal/llm/deepseek"
 	openaiprovider "github.com/iminders/aicoder/internal/llm/openai"
 	"github.com/iminders/aicoder/internal/logger"
+	"github.com/iminders/aicoder/internal/skills"
 	"github.com/iminders/aicoder/internal/slash"
 	"github.com/iminders/aicoder/internal/ui"
 	"github.com/iminders/aicoder/pkg/version"
@@ -25,6 +26,7 @@ import (
 	_ "github.com/iminders/aicoder/internal/tools/filesystem"
 	_ "github.com/iminders/aicoder/internal/tools/search"
 	_ "github.com/iminders/aicoder/internal/tools/shell"
+
 )
 
 // flags holds CLI flag values.
@@ -106,6 +108,11 @@ func Execute() {
 	// Init logger
 	logger.Init(flags.verbose)
 
+	// Load skills (built-ins + user custom)
+	if err := skills.Load(); err != nil {
+		logger.Warn("skill load error: %v", err)
+	}
+
 	// Build provider
 	provider, err := buildProvider(cfg)
 	if err != nil {
@@ -170,88 +177,8 @@ func runOneShot(a *agent.Agent, prompt string) {
 }
 
 func runInteractive(a *agent.Agent, cfg *config.Config) {
-	slashHandler := slash.NewHandler(a.Session(), cfg)
-
-	// Setup readline with tab completion
-	completer := readline.NewPrefixCompleter()
-	for _, cmd := range slash.AllCommands() {
-		completer.Children = append(completer.Children, readline.PcItem(cmd.Name))
-	}
-
-	rl, err := readline.NewEx(&readline.Config{
-		Prompt:          "\033[1;34m> \033[0m",
-		HistoryFile:     getHistoryFile(),
-		AutoComplete:    completer,
-		InterruptPrompt: "^C",
-		EOFPrompt:       "exit",
-	})
-	if err != nil {
-		// Fallback to basic reader if readline fails
-		runInteractiveBasic(a, cfg)
-		return
-	}
-	defer rl.Close()
-
-	// Setup signal handling for Ctrl+C (interrupt current task, not exit)
-	sigCh := make(chan os.Signal, 1)
-	signal.Notify(sigCh, os.Interrupt)
-
-	var cancelCurrent context.CancelFunc
-
-	go func() {
-		for range sigCh {
-			if cancelCurrent != nil {
-				fmt.Println("\n\033[33m[任务已中断，输入新的指令继续]\033[0m")
-				cancelCurrent()
-				cancelCurrent = nil
-			}
-		}
-	}()
-
-	for {
-		line, err := rl.Readline()
-		if err != nil {
-			if err == readline.ErrInterrupt {
-				if cancelCurrent != nil {
-					cancelCurrent()
-					cancelCurrent = nil
-				}
-				continue
-			} else if err == io.EOF {
-				fmt.Println("\n再见！")
-				break
-			}
-			continue
-		}
-		input := strings.TrimSpace(line)
-		if input == "" {
-			continue
-		}
-
-		// Handle slash commands
-		if strings.HasPrefix(input, "/") {
-			handled, shouldExit := slashHandler.Handle(input)
-			if shouldExit {
-				fmt.Println("再见！")
-				return
-			}
-			if handled {
-				continue
-			}
-		}
-
-		// Run agent
-		ctx, cancel := context.WithCancel(context.Background())
-		cancelCurrent = cancel
-
-		ui.PrintDivider()
-		if err := a.Run(ctx, input); err != nil && ctx.Err() == nil {
-			ui.PrintError(err.Error())
-		}
-		cancel()
-		cancelCurrent = nil
-		ui.PrintDivider()
-	}
+	// Use simple interactive mode (bubbletea TUI conflicts with agent's streaming output)
+	runInteractiveBasic(a, cfg)
 }
 
 func isPipeInput() bool {
@@ -276,8 +203,15 @@ func buildProvider(cfg *config.Config) (llm.Provider, error) {
 			return nil, fmt.Errorf("未找到 OpenAI API Key，请设置 OPENAI_API_KEY 环境变量")
 		}
 		return openaiprovider.New(apiKey, cfg.BaseURL, cfg.Model), nil
+	case "deepseek":
+		// For local deployments, API key is optional
+		// If baseURL is set to localhost/127.0.0.1, allow empty API key
+		if apiKey == "" && (cfg.BaseURL == "" || (!strings.Contains(cfg.BaseURL, "localhost") && !strings.Contains(cfg.BaseURL, "127.0.0.1"))) {
+			return nil, fmt.Errorf("未找到 DeepSeek API Key，请设置 DEEPSEEK_API_KEY 环境变量")
+		}
+		return deepseekprovider.New(apiKey, cfg.BaseURL, cfg.Model), nil
 	default:
-		return nil, fmt.Errorf("不支持的 provider: %s (支持: anthropic, openai)", cfg.Provider)
+		return nil, fmt.Errorf("不支持的 provider: %s (支持: anthropic, openai, deepseek)", cfg.Provider)
 	}
 }
 
@@ -315,15 +249,38 @@ func runInteractiveBasic(a *agent.Agent, cfg *config.Config) {
 
 	for {
 		fmt.Print("\033[1;34m> \033[0m")
-		line, err := reader.ReadString('\n')
-		if err != nil {
-			if err == io.EOF {
-				fmt.Println("\n再见！")
+
+		// Read input with multi-line support
+		// Use Ctrl+D (EOF) to submit, or empty line after content
+		var inputLines []string
+		for {
+			line, err := reader.ReadString('\n')
+			if err != nil {
+				if err == io.EOF {
+					if len(inputLines) > 0 {
+						// Submit accumulated input
+						break
+					}
+					fmt.Println("\n再见！")
+					return
+				}
+				continue
+			}
+
+			trimmed := strings.TrimSpace(line)
+
+			// If empty line and we have content, submit
+			if trimmed == "" && len(inputLines) > 0 {
 				break
 			}
-			continue
+
+			// If not empty, accumulate
+			if trimmed != "" {
+				inputLines = append(inputLines, line)
+			}
 		}
-		input := strings.TrimSpace(line)
+
+		input := strings.TrimSpace(strings.Join(inputLines, ""))
 		if input == "" {
 			continue
 		}
@@ -336,6 +293,23 @@ func runInteractiveBasic(a *agent.Agent, cfg *config.Config) {
 				return
 			}
 			if handled {
+				// Check if /skill <n> <prompt> queued a skill run
+				if a.Session().PendingSkillName != "" {
+					skillName := a.Session().PendingSkillName
+					prompt := a.Session().PendingPrompt
+					a.Session().PendingSkillName = ""
+					a.Session().PendingPrompt = ""
+ 
+					ctx, cancel := context.WithCancel(context.Background())
+					cancelCurrent = cancel
+					ui.PrintDivider()
+					if err := a.RunWithSkillByName(ctx, prompt, skillName); err != nil && ctx.Err() == nil {
+						ui.PrintError(err.Error())
+					}
+					cancel()
+					cancelCurrent = nil
+					ui.PrintDivider()
+				}
 				continue
 			}
 		}
diff --git a/docs/PHASE6_7_COMPLETION.md b/docs/PHASE6_7_COMPLETION.md
index 4c03bb7..ffab8c1 100644
--- a/docs/PHASE6_7_COMPLETION.md
+++ b/docs/PHASE6_7_COMPLETION.md
@@ -28,7 +28,7 @@
 | `/cost` | ✅ | 展示 Token 消耗和费用估算 |
 | `/model [name]` | ✅ | 查看或热切换 AI 模型 |
 | `/config [set key value]` | ✅ | 查看或修改配置并持久化 |
-| `/init` | ✅ | 生成 AICODER.md 模板 |
+| `/init` | ✅ | 生成 .AICODER.md 模板 |
 | `/exit`, `/quit`, `/q` | ✅ | 优雅退出程序 |
 
 #### 3. Tab 补全功能 (`internal/slash/completion.go`) 🆕
diff --git a/docs/PHASE6_7_SUMMARY.md b/docs/PHASE6_7_SUMMARY.md
index 3691fb7..d735e36 100644
--- a/docs/PHASE6_7_SUMMARY.md
+++ b/docs/PHASE6_7_SUMMARY.md
@@ -62,8 +62,8 @@ This document summarizes the implementation of Phase 6 (Slash Commands) and Phas
    - Validates values (e.g., provider must be anthropic/openai)
    - Persists changes to user config file
 
-10. **`/init`** - Initialize AICODER.md template
-    - Creates AICODER.md in current directory
+10. **`/init`** - Initialize .AICODER.md template
+    - Creates .AICODER.md in current directory
     - Includes sections: Project Description, Code Standards, Common Commands, Notes
     - Adds version and timestamp footer
 
diff --git a/docs/arch.md b/docs/arch.md
index 7a9d0c4..af486a8 100644
--- a/docs/arch.md
+++ b/docs/arch.md
@@ -104,7 +104,7 @@ aicoder/
 │   │   ├── collector.go           # 上下文收集主入口
 │   │   ├── git.go                 # Git 信息（status/diff/log）
 │   │   ├── project.go             # 项目语言/依赖检测
-│   │   ├── aicoder_md.go          # AICODER.md 加载与解析
+│   │   ├── aicoder_md.go          # .AICODER.md 加载与解析
 │   │   └── summarizer.go          # 目录结构摘要生成
 │   │
 │   ├── session/                   # 会话管理
@@ -659,7 +659,7 @@ type FileSnapshot struct {
 **实现状态：** ✅ 已完成
 
 **收集的信息：**
-- AICODER.md 内容
+- .AICODER.md 内容
 - Git 状态 (分支、修改、最近提交)
 - 项目类型检测 (Go/Node.js/Python/Rust/Java/Ruby)
 - 项目根目录
@@ -667,7 +667,7 @@ type FileSnapshot struct {
 **系统提示词构成：**
 ```
 基础角色定义
-+ AICODER.md (项目说明)
++ .AICODER.md (项目说明)
 + 项目环境信息
 + Git 状态
 ```
@@ -753,7 +753,7 @@ type FileSnapshot struct {
 **待实现：**
 - ⏳ 多 Agent 并行任务
 - ⏳ Web Dashboard
-- ⏳ AICODER.md 模板市场
+- ⏳ .AICODER.md 模板市场
 - ⏳ VS Code 插件
 
 ### 8.4 技术债务和改进空间
diff --git a/docs/code-guide.md b/docs/code-guide.md
index 34452d6..594d729 100644
--- a/docs/code-guide.md
+++ b/docs/code-guide.md
@@ -488,7 +488,7 @@ func Collect() (*ProjectContext, error)
 
 收集内容:
 1. 查找项目根目录
-2. 读取 AICODER.md
+2. 读取 .AICODER.md
 3. 收集 Git 信息
 4. 检测项目类型
 
@@ -500,7 +500,7 @@ func (pc *ProjectContext) BuildSystemPrompt() string
 
 组合:
 - 基础角色定义
-- AICODER.md 内容
+- .AICODER.md 内容
 - 项目环境信息
 - Git 状态
 
diff --git a/docs/prd.md b/docs/prd.md
index f74dc8b..74e6c25 100644
--- a/docs/prd.md
+++ b/docs/prd.md
@@ -130,16 +130,16 @@ AI 能够调用一组内置工具来自主完成任务，工具调用前须向
 | `/cost` | 查看当前会话已消耗的 Token 及费用估算 |
 | `/model` | 切换使用的 AI 模型 |
 | `/config` | 查看或修改配置项 |
-| `/init` | 在当前项目初始化 `aicoder` 配置（生成 `AICODER.md`） |
+| `/init` | 在当前项目初始化 `aicoder` 配置（生成 `.AICODER.md`） |
 | `/exit` | 退出程序 |
 
 ---
 
 ### 2.4 项目上下文感知（Project Context）
 
-#### 2.4.1 AICODER.md
+#### 2.4.1 .AICODER.md
 
-项目根目录下的 `AICODER.md` 文件作为**项目级系统提示词**，AI 在每次会话开始时自动加载：
+项目根目录下的 `.AICODER.md` 文件作为**项目级系统提示词**，AI 在每次会话开始时自动加载：
 
 ```markdown
 # 项目说明
@@ -394,7 +394,7 @@ go install github.com/iminders/aicoder@latest
 
 - [ ] 多 Agent 并行任务
 - [ ] 可视化 Web Dashboard（会话历史、费用分析）
-- [ ] 团队共享 `AICODER.md` 模板市场
+- [ ] 团队共享 `.AICODER.md` 模板市场
 - [ ] IDE 插件（VS Code）集成本 CLI
 
 ---
diff --git a/docs/todo.md b/docs/todo.md
index 2f8c217..942232c 100644
--- a/docs/todo.md
+++ b/docs/todo.md
@@ -339,7 +339,7 @@
 
 - [ ] 多 Agent 并行任务（任务拆解 + 子 Agent 协同）
 - [ ] Web Dashboard（会话历史可视化、Token 用量分析）
-- [ ] `AICODER.md` 模板市场（团队共享最佳实践）
+- [ ] `.AICODER.md` 模板市场（团队共享最佳实践）
 - [ ] VS Code 插件（内嵌本 CLI，提供侧边栏 UI）
 - [ ] 企业版：SSO 集成、操作审计日志导出、私有部署
 
diff --git a/go.mod b/go.mod
index 69dd7c6..4ae9283 100644
--- a/go.mod
+++ b/go.mod
@@ -1,12 +1,12 @@
 module github.com/iminders/aicoder
 
-go 1.19
+go 1.24.0
 
 require (
 	github.com/charmbracelet/bubbletea v0.25.0
 	github.com/charmbracelet/glamour v0.6.0
 	github.com/charmbracelet/lipgloss v0.9.1
-	github.com/chzyer/readline v1.5.1
+	github.com/mattn/go-isatty v0.0.20
 )
 
 require (
@@ -17,21 +17,20 @@ require (
 	github.com/dlclark/regexp2 v1.4.0 // indirect
 	github.com/gorilla/css v1.0.0 // indirect
 	github.com/lucasb-eyer/go-colorful v1.2.0 // indirect
-	github.com/mattn/go-isatty v0.0.18 // indirect
 	github.com/mattn/go-localereader v0.0.1 // indirect
-	github.com/mattn/go-runewidth v0.0.15 // indirect
+	github.com/mattn/go-runewidth v0.0.16 // indirect
 	github.com/microcosm-cc/bluemonday v1.0.21 // indirect
-	github.com/muesli/ansi v0.0.0-20211018074035-2e021307bc4b // indirect
+	github.com/muesli/ansi v0.0.0-20230316100256-276c6243b2f6 // indirect
 	github.com/muesli/cancelreader v0.2.2 // indirect
 	github.com/muesli/reflow v0.3.0 // indirect
-	github.com/muesli/termenv v0.15.2 // indirect
+	github.com/muesli/termenv v0.16.0 // indirect
 	github.com/olekukonko/tablewriter v0.0.5 // indirect
-	github.com/rivo/uniseg v0.2.0 // indirect
+	github.com/rivo/uniseg v0.4.7 // indirect
 	github.com/yuin/goldmark v1.5.2 // indirect
 	github.com/yuin/goldmark-emoji v1.0.1 // indirect
 	golang.org/x/net v0.0.0-20221002022538-bcab6841153b // indirect
-	golang.org/x/sync v0.1.0 // indirect
-	golang.org/x/sys v0.12.0 // indirect
+	golang.org/x/sync v0.8.0 // indirect
+	golang.org/x/sys v0.36.0 // indirect
 	golang.org/x/term v0.6.0 // indirect
 	golang.org/x/text v0.3.8 // indirect
 )
diff --git a/go.sum b/go.sum
index bac4741..ae54981 100644
--- a/go.sum
+++ b/go.sum
@@ -11,12 +11,6 @@ github.com/charmbracelet/glamour v0.6.0 h1:wi8fse3Y7nfcabbbDuwolqTqMQPMnVPeZhDM2
 github.com/charmbracelet/glamour v0.6.0/go.mod h1:taqWV4swIMMbWALc0m7AfE9JkPSU8om2538k9ITBxOc=
 github.com/charmbracelet/lipgloss v0.9.1 h1:PNyd3jvaJbg4jRHKWXnCj1akQm4rh8dbEzN1p/u1KWg=
 github.com/charmbracelet/lipgloss v0.9.1/go.mod h1:1mPmG4cxScwUQALAAnacHaigiiHB9Pmr+v1VEawJl6I=
-github.com/chzyer/logex v1.2.1 h1:XHDu3E6q+gdHgsdTPH6ImJMIp436vR6MPtH8gP05QzM=
-github.com/chzyer/logex v1.2.1/go.mod h1:JLbx6lG2kDbNRFnfkgvh4eRJRPX1QCoOIWomwysCBrQ=
-github.com/chzyer/readline v1.5.1 h1:upd/6fQk4src78LMRzh5vItIt361/o4uq553V8B5sGI=
-github.com/chzyer/readline v1.5.1/go.mod h1:Eh+b79XXUwfKfcPLepksvw2tcLE/Ct21YObkaSkeBlk=
-github.com/chzyer/test v1.0.0 h1:p3BQDXSxOhOG0P9z6/hGnII4LGiEPOYBhs8asl/fC04=
-github.com/chzyer/test v1.0.0/go.mod h1:2JlltgoNkt4TW/z9V/IzDdFaMTM2JPIi26O1pF38GC8=
 github.com/containerd/console v1.0.4-0.20230313162750-1ae8d489ac81 h1:q2hJAaP1k2wIvVRd/hEHD7lacgqrCPS+k8g1MndzfWY=
 github.com/containerd/console v1.0.4-0.20230313162750-1ae8d489ac81/go.mod h1:YynlIjWYF8myEu6sdkwKIvGQq+cOckRm6So2avqoYAk=
 github.com/davecgh/go-spew v1.1.0/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
@@ -29,33 +23,34 @@ github.com/gorilla/css v1.0.0/go.mod h1:Dn721qIggHpt4+EFCcTLTU/vk5ySda2ReITrtgBl
 github.com/lucasb-eyer/go-colorful v1.2.0 h1:1nnpGOrhyZZuNyfu1QjKiUICQ74+3FNCN69Aj6K7nkY=
 github.com/lucasb-eyer/go-colorful v1.2.0/go.mod h1:R4dSotOR9KMtayYi1e77YzuveK+i7ruzyGqttikkLy0=
 github.com/mattn/go-isatty v0.0.16/go.mod h1:kYGgaQfpe5nmfYZH+SKPsOc2e4SrIfOl2e/yFXSvRLM=
-github.com/mattn/go-isatty v0.0.18 h1:DOKFKCQ7FNG2L1rbrmstDN4QVRdS89Nkh85u68Uwp98=
-github.com/mattn/go-isatty v0.0.18/go.mod h1:W+V8PltTTMOvKvAeJH7IuucS94S2C6jfK/D7dTCTo3Y=
+github.com/mattn/go-isatty v0.0.20 h1:xfD0iDuEKnDkl03q4limB+vH+GxLEtL/jb4xVJSWWEY=
+github.com/mattn/go-isatty v0.0.20/go.mod h1:W+V8PltTTMOvKvAeJH7IuucS94S2C6jfK/D7dTCTo3Y=
 github.com/mattn/go-localereader v0.0.1 h1:ygSAOl7ZXTx4RdPYinUpg6W99U8jWvWi9Ye2JC/oIi4=
 github.com/mattn/go-localereader v0.0.1/go.mod h1:8fBrzywKY7BI3czFoHkuzRoWE9C+EiG4R1k4Cjx5p88=
 github.com/mattn/go-runewidth v0.0.9/go.mod h1:H031xJmbD/WCDINGzjvQ9THkh0rPKHF+m2gUSrubnMI=
 github.com/mattn/go-runewidth v0.0.12/go.mod h1:RAqKPSqVFrSLVXbA8x7dzmKdmGzieGRCM46jaSJTDAk=
 github.com/mattn/go-runewidth v0.0.14/go.mod h1:Jdepj2loyihRzMpdS35Xk/zdY8IAYHsh153qUoGf23w=
-github.com/mattn/go-runewidth v0.0.15 h1:UNAjwbU9l54TA3KzvqLGxwWjHmMgBUVhBiTjelZgg3U=
-github.com/mattn/go-runewidth v0.0.15/go.mod h1:Jdepj2loyihRzMpdS35Xk/zdY8IAYHsh153qUoGf23w=
+github.com/mattn/go-runewidth v0.0.16 h1:E5ScNMtiwvlvB5paMFdw9p4kSQzbXFikJ5SQO6TULQc=
+github.com/mattn/go-runewidth v0.0.16/go.mod h1:Jdepj2loyihRzMpdS35Xk/zdY8IAYHsh153qUoGf23w=
 github.com/microcosm-cc/bluemonday v1.0.21 h1:dNH3e4PSyE4vNX+KlRGHT5KrSvjeUkoNPwEORjffHJg=
 github.com/microcosm-cc/bluemonday v1.0.21/go.mod h1:ytNkv4RrDrLJ2pqlsSI46O6IVXmZOBBD4SaJyDwwTkM=
-github.com/muesli/ansi v0.0.0-20211018074035-2e021307bc4b h1:1XF24mVaiu7u+CFywTdcDo2ie1pzzhwjt6RHqzpMU34=
-github.com/muesli/ansi v0.0.0-20211018074035-2e021307bc4b/go.mod h1:fQuZ0gauxyBcmsdE3ZT4NasjaRdxmbCS0jRHsrWu3Ho=
+github.com/muesli/ansi v0.0.0-20230316100256-276c6243b2f6 h1:ZK8zHtRHOkbHy6Mmr5D264iyp3TiX5OmNcI5cIARiQI=
+github.com/muesli/ansi v0.0.0-20230316100256-276c6243b2f6/go.mod h1:CJlz5H+gyd6CUWT45Oy4q24RdLyn7Md9Vj2/ldJBSIo=
 github.com/muesli/cancelreader v0.2.2 h1:3I4Kt4BQjOR54NavqnDogx/MIoWBFa0StPA8ELUXHmA=
 github.com/muesli/cancelreader v0.2.2/go.mod h1:3XuTXfFS2VjM+HTLZY9Ak0l6eUKfijIfMUZ4EgX0QYo=
 github.com/muesli/reflow v0.3.0 h1:IFsN6K9NfGtjeggFP+68I4chLZV2yIKsXJFNZ+eWh6s=
 github.com/muesli/reflow v0.3.0/go.mod h1:pbwTDkVPibjO2kyvBQRBxTWEEGDGq0FlB1BIKtnHY/8=
 github.com/muesli/termenv v0.13.0/go.mod h1:sP1+uffeLaEYpyOTb8pLCUctGcGLnoFjSn4YJK5e2bc=
-github.com/muesli/termenv v0.15.2 h1:GohcuySI0QmI3wN8Ok9PtKGkgkFIk7y6Vpb5PvrY+Wo=
-github.com/muesli/termenv v0.15.2/go.mod h1:Epx+iuz8sNs7mNKhxzH4fWXGNpZwUaJKRS1noLXviQ8=
+github.com/muesli/termenv v0.16.0 h1:S5AlUN9dENB57rsbnkPyfdGuWIlkmzJjbFf0Tf5FWUc=
+github.com/muesli/termenv v0.16.0/go.mod h1:ZRfOIKPFDYQoDFF4Olj7/QJbW60Ol/kL1pU3VfY/Cnk=
 github.com/olekukonko/tablewriter v0.0.5 h1:P2Ga83D34wi1o9J6Wh1mRuqd4mF/x/lgBS7N7AbDhec=
 github.com/olekukonko/tablewriter v0.0.5/go.mod h1:hPp6KlRPjbx+hW8ykQs1w3UBbZlj6HuIJcUGPhkA7kY=
 github.com/pmezard/go-difflib v1.0.0 h1:4DBwDE0NGyQoBHbLQYPwSUPoCMWR5BEzIk/f1lZbAQM=
 github.com/pmezard/go-difflib v1.0.0/go.mod h1:iKH77koFhYxTK1pcRnkKkqfTogsbg7gZNVY4sRDYZ/4=
 github.com/rivo/uniseg v0.1.0/go.mod h1:J6wj4VEh+S6ZtnVlnTBMWIodfgj8LQOQFoIToxlJtxc=
-github.com/rivo/uniseg v0.2.0 h1:S1pD9weZBuJdFmowNwbpi7BJ8TNftyUImj/0WQi72jY=
 github.com/rivo/uniseg v0.2.0/go.mod h1:J6wj4VEh+S6ZtnVlnTBMWIodfgj8LQOQFoIToxlJtxc=
+github.com/rivo/uniseg v0.4.7 h1:WUdvkW8uEhrYfLC4ZzdpI2ztxP1I582+49Oc5Mq64VQ=
+github.com/rivo/uniseg v0.4.7/go.mod h1:FN3SvrM+Zdj16jyLfmOkMNblXMcoc8DfTHruCPUcx88=
 github.com/stretchr/objx v0.1.0/go.mod h1:HFkY916IF+rwdDfMAkV7OtwuqBVzrE8GR6GFx+wExME=
 github.com/stretchr/testify v1.7.0 h1:nwc3DEeHmmLAfoZucVR881uASk0Mfjw8xYJ99tb5CcY=
 github.com/stretchr/testify v1.7.0/go.mod h1:6Fq8oRcR53rry900zMqJjRRixrwX3KX962/h/Wwjteg=
@@ -66,16 +61,15 @@ github.com/yuin/goldmark-emoji v1.0.1 h1:ctuWEyzGBwiucEqxzwe0SOYDXPAucOrE9NQC18W
 github.com/yuin/goldmark-emoji v1.0.1/go.mod h1:2w1E6FEWLcDQkoTE+7HU6QF1F6SLlNGjRIBbIZQFqkQ=
 golang.org/x/net v0.0.0-20221002022538-bcab6841153b h1:6e93nYa3hNqAvLr0pD4PN1fFS+gKzp2zAXqrnTCstqU=
 golang.org/x/net v0.0.0-20221002022538-bcab6841153b/go.mod h1:YDH+HFinaLZZlnHAfSS6ZXJJ9M9t4Dl22yv3iI2vPwk=
-golang.org/x/sync v0.1.0 h1:wsuoTGHzEhffawBOhz5CYhcrV4IdKZbEyZjBMuTp12o=
-golang.org/x/sync v0.1.0/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
+golang.org/x/sync v0.8.0 h1:3NFvSEYkUoMifnESzZl15y791HH1qU2xm6eCJU5ZPXQ=
+golang.org/x/sync v0.8.0/go.mod h1:Czt+wKu1gCyEFDUtn0jG5QVvpJ6rzVqr5aXyt9drQfk=
 golang.org/x/sys v0.0.0-20210615035016-665e8c7367d1/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
-golang.org/x/sys v0.0.0-20220310020820-b874c991c1a5/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.0.0-20220728004956-3c1f35247d10/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.0.0-20220811171246-fbc7d0a398ab/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.1.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.6.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
-golang.org/x/sys v0.12.0 h1:CM0HF96J0hcLAwsHPJZjfdNzs0gftsLfgKt57wWHJ0o=
-golang.org/x/sys v0.12.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
+golang.org/x/sys v0.36.0 h1:KVRy2GtZBrk1cBYA7MKu5bEZFxQk4NIDV6RLVcC8o0k=
+golang.org/x/sys v0.36.0/go.mod h1:OgkHotnGiDImocRcuBABYBEXf8A9a87e/uXjp9XT3ks=
 golang.org/x/term v0.0.0-20210927222741-03fcf44c2211/go.mod h1:jbD1KX2456YbFQfuXm/mYQcufACuNUgVhRMnK/tPxf8=
 golang.org/x/term v0.6.0 h1:clScbb1cHjoCkyRbWwBEUZ5H/tIFu5TAXIqaZD0Gcjw=
 golang.org/x/term v0.6.0/go.mod h1:m6U89DPEgQRMq3DNkDClhWw02AUbt2daBVO4cn4Hv9U=
diff --git a/internal/agent/agent.go b/internal/agent/agent.go
index 890bbec..e6f6790 100644
--- a/internal/agent/agent.go
+++ b/internal/agent/agent.go
@@ -3,6 +3,7 @@ package agent
 import (
 	"bufio"
 	"context"
+	"encoding/json"
 	"fmt"
 	"os"
 	"strings"
@@ -13,8 +14,10 @@ import (
 	"github.com/iminders/aicoder/internal/llm"
 	"github.com/iminders/aicoder/internal/logger"
 	"github.com/iminders/aicoder/internal/session"
+	"github.com/iminders/aicoder/internal/skills"
 	"github.com/iminders/aicoder/internal/tools"
 	"github.com/iminders/aicoder/internal/ui"
+	"github.com/mattn/go-isatty"
 )
 
 // Agent orchestrates the full LLM ↔ tool loop.
@@ -55,10 +58,45 @@ func New(cfg *config.Config, provider llm.Provider) *Agent {
 func (a *Agent) Session() *session.Session { return a.sess }
 
 // Run processes one user turn and drives the Agent Loop until the model stops.
+// It auto-detects applicable skills and injects their prompts.
 func (a *Agent) Run(ctx context.Context, userInput string) error {
+	// Auto-detect skill from input
+	if skill := skills.Global.Detect(userInput); skill != nil {
+		return a.RunWithSkill(ctx, userInput, skill)
+	}
+	return a.run(ctx, userInput)
+}
+
+// RunWithSkill runs with a specific skill injected as additional context.
+func (a *Agent) RunWithSkill(ctx context.Context, userInput string, skill *skills.Skill) error {
+	ui.PrintInfo(fmt.Sprintf("🎯 技能激活: %s — %s", skill.Name, skill.Description))
+	if skill.OutputFile != "" {
+		ui.PrintInfo(fmt.Sprintf("📄 建议输出文件: %s", skill.OutputFile))
+	}
+
+	// Inject skill prompt as a system-level context message for this turn only.
+	// We add it as a user message prefix so it doesn't pollute the system prompt.
+	augmented := fmt.Sprintf("[技能上下文: %s]\n%s\n\n---\n用户需求：%s",
+		skill.Name, skill.Prompt, userInput)
+	return a.run(ctx, augmented)
+}
+
+// RunWithSkillByName looks up a skill by name and delegates to RunWithSkill.
+func (a *Agent) RunWithSkillByName(ctx context.Context, userInput, skillName string) error {
+	skill := skills.Global.Get(skillName)
+	if skill == nil {
+		// Skill not found — run without it but warn
+		ui.PrintWarn(fmt.Sprintf("Skill %q not found, running without skill context", skillName))
+		return a.run(ctx, userInput)
+	}
+	return a.RunWithSkill(ctx, userInput, skill)
+}
+
+// Run processes one user turn and drives the Agent Loop until the model stops.
+func (a *Agent) run(ctx context.Context, userInput string) error {
 	a.sess.AppendMessage(session.TextMessage("user", userInput))
 
-	for iteration := 0; iteration < 20; iteration++ {
+	for iteration := 0; iteration < 50; iteration++ {
 		// Build the LLM request
 		req := &llm.Request{
 			Model:     a.cfg.Model,
@@ -76,6 +114,11 @@ func (a *Agent) Run(ctx context.Context, userInput string) error {
 		// Consume the stream
 		var toolCalls []llm.ToolUseBlock
 		var textBuf strings.Builder
+		var thinkingActive bool
+		var thinkingChars int
+
+		// Check if stdout is a terminal
+		isTTY := isatty.IsTerminal(os.Stdout.Fd())
 
 		for event := range eventCh {
 			select {
@@ -86,8 +129,33 @@ func (a *Agent) Run(ctx context.Context, userInput string) error {
 			}
 			switch event.Type {
 			case "text_delta":
+				// If we were thinking, clear the thinking indicator
+				if thinkingActive && isTTY {
+					fmt.Print("\r\033[K") // Clear line
+				}
+				thinkingActive = false
+				thinkingChars = 0
 				a.renderer.Write(event.Delta)
 				textBuf.WriteString(event.Delta)
+			case "thinking_delta":
+				// Show thinking progress without newline (only in TTY mode)
+				thinkingChars += len(event.Delta)
+				if !thinkingActive {
+					thinkingActive = true
+				}
+				// Only update indicator in TTY mode to avoid cluttering pipe output
+				if isTTY {
+					fmt.Printf("\r\033[90m[Thinking... %d chars]\033[0m", thinkingChars)
+				}
+			case "thinking_done":
+				// DeepSeek R1 thinking complete (</think> detected)
+				if thinkingActive && isTTY {
+					fmt.Print("\r\033[K") // Clear the thinking line
+				}
+				thinkingActive = false
+				thinkingChars = 0
+				logger.Debug("thinking complete: %d chars", len(event.Delta))
+				// Don't write thinking content to output
 			case "tool_use_end":
 				if event.ToolUse != nil {
 					toolCalls = append(toolCalls, *event.ToolUse)
@@ -163,6 +231,8 @@ func (a *Agent) executeToolCall(ctx context.Context, tc llm.ToolUseBlock) sessio
 
 	// Execute
 	fmt.Printf("\033[36m⚙  执行 %s...\033[0m\n", tc.Name)
+	printToolCallPath(tc.Input)
+
 	start := time.Now()
 	result, err := tool.Execute(ctx, tc.Input)
 	elapsed := time.Since(start)
@@ -186,6 +256,27 @@ func (a *Agent) executeToolCall(ctx context.Context, tc llm.ToolUseBlock) sessio
 	}
 }
 
+func printToolCallPath(raw json.RawMessage) {
+	var result struct {
+		Path    *string `json:"path"`
+		Command *string `json:"command"`
+	}
+
+	if err := json.Unmarshal(raw, &result); err != nil {
+		fmt.Println("解析失败")
+		return
+	}
+
+	if result.Path != nil {
+		fmt.Printf("%s\n", *result.Path)
+	}
+
+	if result.Command != nil {
+		fmt.Printf("$ %s\n", *result.Command)
+	}
+
+}
+
 func buildAssistantMessage(text string, toolCalls []llm.ToolUseBlock) session.Message {
 	msg := session.Message{Role: "assistant"}
 	if text != "" {
diff --git a/internal/config/config.go b/internal/config/config.go
index fa943ad..769eb00 100644
--- a/internal/config/config.go
+++ b/internal/config/config.go
@@ -136,6 +136,8 @@ func APIKey(provider string) string {
 		return os.Getenv("ANTHROPIC_API_KEY")
 	case "openai":
 		return os.Getenv("OPENAI_API_KEY")
+	case "deepseek":
+		return os.Getenv("DEEPSEEK_API_KEY")
 	case "google":
 		return os.Getenv("GOOGLE_API_KEY")
 	default:
diff --git a/internal/context/PHASE4_SUMMARY.md b/internal/context/PHASE4_SUMMARY.md
index b9ad7b2..5e3627d 100644
--- a/internal/context/PHASE4_SUMMARY.md
+++ b/internal/context/PHASE4_SUMMARY.md
@@ -131,7 +131,7 @@ aicoder/
 ### 项目上下文收集 (`internal/context/`)
 - [x] `git.go` - Git 信息采集 ⭐
 - [x] `project.go` - 项目类型检测 ⭐
-- [x] `aicoder_md.go` - AICODER.md 加载 ⭐
+- [x] `aicoder_md.go` - .AICODER.md 加载 ⭐
 - [x] `summarizer.go` - 目录结构摘要 ⭐ **新增**
 - [x] `collector.go` - 上下文组合和系统提示生成 ⭐
 
diff --git a/internal/context/collector.go b/internal/context/collector.go
index 5efb945..1a3dcc6 100644
--- a/internal/context/collector.go
+++ b/internal/context/collector.go
@@ -11,7 +11,7 @@ import (
 // ProjectContext holds auto-discovered project information.
 type ProjectContext struct {
 	RootDir       string
-	AICoderMD     string // contents of AICODER.md if found
+	AICoderMD     string // contents of .AICODER.md if found
 	GitInfo       string
 	ProjectInfo   string
 	DirectoryTree string // directory structure summary
@@ -55,7 +55,7 @@ func (c *ProjectContext) SystemPrompt() string {
 func findAICoderMD(startDir string) string {
 	dir := startDir
 	for {
-		path := filepath.Join(dir, "AICODER.md")
+		path := filepath.Join(dir, ".AICODER.md")
 		if data, err := os.ReadFile(path); err == nil {
 			return string(data)
 		}
diff --git a/internal/context/summarizer.go b/internal/context/summarizer.go
index ac8d1af..784eb45 100644
--- a/internal/context/summarizer.go
+++ b/internal/context/summarizer.go
@@ -77,7 +77,8 @@ func SummarizeDirectory(rootPath string, config *SummarizerConfig) string {
 
 	// 渲染为字符串
 	var sb strings.Builder
-	sb.WriteString("项目结构:\n")
+	// sb.WriteString("项目结构(当前根目录为" + filepath.Base(rootPath) + "):\n")
+	sb.WriteString("项目名称为：" + filepath.Base(rootPath) + "，项目结构:\n")
 	totalFiles := 0
 	renderTree(&sb, root, "", true, &totalFiles)
 
@@ -245,10 +246,11 @@ func renderTree(sb *strings.Builder, node *DirectoryNode, prefix string, isLast
 				*totalFiles++
 			}
 		}
-	} else {
-		// 根节点
-		sb.WriteString(node.Name + "/\n")
 	}
+	// else {
+	// 根节点
+	// sb.WriteString(node.Name + "/\n")
+	// }
 
 	// 递归渲染子节点
 	for i, child := range node.Children {
@@ -310,7 +312,7 @@ func GetImportantFiles(rootPath string) []string {
 		"Gemfile",
 		".gitignore",
 		".env.example",
-		"AICODER.md",
+		".AICODER.md",
 	}
 
 	var files []string
diff --git a/internal/llm/deepseek/provider.go b/internal/llm/deepseek/provider.go
new file mode 100644
index 0000000..5ad9e5e
--- /dev/null
+++ b/internal/llm/deepseek/provider.go
@@ -0,0 +1,542 @@
+package deepseek
+
+import (
+	"bufio"
+	"bytes"
+	"context"
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"strings"
+	"time"
+
+	"github.com/iminders/aicoder/internal/llm"
+	"github.com/iminders/aicoder/internal/logger"
+	"github.com/iminders/aicoder/internal/session"
+)
+
+// Provider implements the LLM provider interface for DeepSeek API.
+// DeepSeek R1 uses OpenAI-compatible API but has special handling for:
+// 1. Reasoning tokens wrapped in <think>...</think> tags
+// 2. Token usage may not be provided in streaming responses
+type Provider struct {
+	apiKey  string
+	baseURL string
+	model   string
+	client  *http.Client
+}
+
+// New creates a new DeepSeek provider.
+func New(apiKey, baseURL, model string) *Provider {
+	if baseURL == "" {
+		baseURL = "https://api.deepseek.com"
+	}
+	if model == "" {
+		model = "deepseek-reasoner"
+	}
+	return &Provider{
+		apiKey:  apiKey,
+		baseURL: baseURL,
+		model:   model,
+		client:  &http.Client{Timeout: 3000 * time.Second},
+	}
+}
+
+// Stream sends a streaming request to DeepSeek API.
+func (p *Provider) Stream(ctx context.Context, req *llm.Request) (<-chan llm.StreamEvent, error) {
+	// Type assert messages from RawMsgs
+	messages, ok := req.RawMsgs.([]session.Message)
+	if !ok {
+		return nil, fmt.Errorf("deepseek provider expects []session.Message")
+	}
+
+	apiMessages := convertMessages(messages)
+
+	payload := map[string]interface{}{
+		"model":       p.model,
+		"messages":    apiMessages,
+		"stream":      true,
+		"max_tokens":  req.MaxTokens,
+		"temperature": 1.0,
+	}
+
+	// Add tools if provided
+	if len(req.Tools) > 0 {
+		payload["tools"] = convertTools(req.Tools)
+	}
+
+	body, err := json.Marshal(payload)
+	if err != nil {
+		return nil, fmt.Errorf("marshal request: %w", err)
+	}
+
+	httpReq, err := http.NewRequestWithContext(ctx, "POST", p.baseURL+"/v1/chat/completions", bytes.NewReader(body))
+	if err != nil {
+		return nil, fmt.Errorf("create request: %w", err)
+	}
+
+	logger.Debug("DeepSeek request to: %s", p.baseURL+"/v1/chat/completions")
+	logger.Debug("DeepSeek model: %s", p.model)
+
+	httpReq.Header.Set("Content-Type", "application/json")
+	// Only set Authorization header if API key is provided (not needed for local deployments)
+	if p.apiKey != "" {
+		httpReq.Header.Set("Authorization", "Bearer "+p.apiKey)
+		logger.Debug("DeepSeek: using API key authentication")
+	} else {
+		logger.Debug("DeepSeek: no API key (local deployment mode)")
+	}
+
+	resp, err := p.client.Do(httpReq)
+	if err != nil {
+		return nil, fmt.Errorf("http request: %w", err)
+	}
+
+	if resp.StatusCode != http.StatusOK {
+		body, _ := io.ReadAll(resp.Body)
+		resp.Body.Close()
+		logger.Debug("DeepSeek API error response: %s", string(body))
+		return nil, fmt.Errorf("API error %d: %s", resp.StatusCode, string(body))
+	}
+
+	ch := make(chan llm.StreamEvent, 10)
+
+	go p.readStream(resp.Body, ch, messages)
+
+	return ch, nil
+}
+
+// readStream reads SSE stream from DeepSeek API.
+func (p *Provider) readStream(body io.ReadCloser, ch chan<- llm.StreamEvent, messages []session.Message) {
+	defer close(ch)
+	defer body.Close()
+
+	scanner := bufio.NewScanner(body)
+	scanner.Buffer(make([]byte, 0, 512*1024), 512*1024)
+
+	var textBuffer strings.Builder
+	var toolCalls []llm.ToolUseBlock
+	var inThinking bool
+	var thinkingBuffer strings.Builder
+	var usageReceived bool
+
+	for scanner.Scan() {
+		line := scanner.Text()
+		if !strings.HasPrefix(line, "data: ") {
+			continue
+		}
+
+		data := strings.TrimPrefix(line, "data: ")
+		if data == "[DONE]" {
+			break
+		}
+
+		var chunk struct {
+			Choices []struct {
+				Delta struct {
+					Content   string `json:"content"`
+					ToolCalls []struct {
+						Index    int    `json:"index"`
+						ID       string `json:"id"`
+						Type     string `json:"type"`
+						Function struct {
+							Name      string `json:"name"`
+							Arguments string `json:"arguments"`
+						} `json:"function"`
+					} `json:"tool_calls"`
+				} `json:"delta"`
+				FinishReason string `json:"finish_reason"`
+			} `json:"choices"`
+			Usage *struct {
+				PromptTokens     int `json:"prompt_tokens"`
+				CompletionTokens int `json:"completion_tokens"`
+			} `json:"usage"`
+		}
+
+		if err := json.Unmarshal([]byte(data), &chunk); err != nil {
+			logger.Debug("parse chunk error: %v", err)
+			continue
+		}
+
+		// Handle usage if provided
+		if chunk.Usage != nil {
+			usageReceived = true
+			ch <- llm.StreamEvent{
+				Type:   "usage",
+				Input:  chunk.Usage.PromptTokens,
+				Output: chunk.Usage.CompletionTokens,
+			}
+		}
+
+		if len(chunk.Choices) == 0 {
+			continue
+		}
+
+		choice := chunk.Choices[0]
+
+		// Handle text content with thinking detection
+		if choice.Delta.Content != "" {
+			content := choice.Delta.Content
+
+			// Check for thinking tags
+			if strings.Contains(content, "<think>") {
+				inThinking = true
+				// Find where <think> starts and split content
+				idx := strings.Index(content, "<think>")
+				if idx > 0 {
+					// Emit content before <think>
+					beforeThink := content[:idx]
+					textBuffer.WriteString(beforeThink)
+					ch <- llm.StreamEvent{
+						Type:  "text_delta",
+						Delta: beforeThink,
+					}
+				}
+				// Start collecting thinking content (skip the <think> tag itself)
+				afterTag := content[idx+7:] // len("<think>") = 7
+
+				// Check if </think> is also in this chunk
+				if strings.Contains(afterTag, "</think>") {
+					endIdx := strings.Index(afterTag, "</think>")
+					thinkingContent := afterTag[:endIdx]
+					thinkingBuffer.WriteString(thinkingContent)
+
+					// Emit thinking_delta for the content
+					if len(thinkingContent) > 0 {
+						ch <- llm.StreamEvent{
+							Type:  "thinking_delta",
+							Delta: thinkingContent,
+						}
+					}
+
+					// Immediately end thinking
+					inThinking = false
+					ch <- llm.StreamEvent{
+						Type:  "thinking_done",
+						Delta: thinkingBuffer.String(),
+					}
+					thinkingBuffer.Reset()
+
+					// Process content after </think>
+					afterThink := afterTag[endIdx+8:] // len("</think>") = 8
+					if len(afterThink) > 0 {
+						textBuffer.WriteString(afterThink)
+						ch <- llm.StreamEvent{
+							Type:  "text_delta",
+							Delta: afterThink,
+						}
+					}
+				} else {
+					// No </think> yet, just accumulate
+					thinkingBuffer.WriteString(afterTag)
+					if len(afterTag) > 0 {
+						ch <- llm.StreamEvent{
+							Type:  "thinking_delta",
+							Delta: afterTag,
+						}
+					}
+				}
+				continue
+			}
+
+			if inThinking {
+				// Check if this chunk contains </think>
+				if strings.Contains(content, "</think>") {
+					idx := strings.Index(content, "</think>")
+					// Add content before </think> to thinking buffer
+					beforeEnd := content[:idx]
+					if len(beforeEnd) > 0 {
+						thinkingBuffer.WriteString(beforeEnd)
+						ch <- llm.StreamEvent{
+							Type:  "thinking_delta",
+							Delta: beforeEnd,
+						}
+					}
+
+					// End thinking immediately
+					inThinking = false
+					ch <- llm.StreamEvent{
+						Type:  "thinking_done",
+						Delta: thinkingBuffer.String(),
+					}
+					thinkingBuffer.Reset()
+
+					// Process content after </think>
+					afterThink := content[idx+8:] // len("</think>") = 8
+					if len(afterThink) > 0 {
+						textBuffer.WriteString(afterThink)
+						ch <- llm.StreamEvent{
+							Type:  "text_delta",
+							Delta: afterThink,
+						}
+					}
+				} else {
+					// Still thinking, accumulate
+					thinkingBuffer.WriteString(content)
+					ch <- llm.StreamEvent{
+						Type:  "thinking_delta",
+						Delta: content,
+					}
+				}
+			} else {
+				// Normal content, not thinking
+				textBuffer.WriteString(content)
+				ch <- llm.StreamEvent{
+					Type:  "text_delta",
+					Delta: content,
+				}
+			}
+		}
+
+		// Handle tool calls
+		if len(choice.Delta.ToolCalls) > 0 {
+			for _, tc := range choice.Delta.ToolCalls {
+				if tc.Index >= len(toolCalls) {
+					// Initialize new tool call
+					inputJSON, _ := json.Marshal(map[string]string{"_args": ""})
+					toolCalls = append(toolCalls, llm.ToolUseBlock{
+						ID:    tc.ID,
+						Name:  tc.Function.Name,
+						Input: inputJSON,
+					})
+				}
+				if tc.Function.Arguments != "" {
+					// Accumulate arguments as JSON string
+					var currentInput map[string]string
+					json.Unmarshal(toolCalls[tc.Index].Input, &currentInput)
+					if currentInput == nil {
+						currentInput = map[string]string{}
+					}
+					currentInput["_args"] += tc.Function.Arguments
+					toolCalls[tc.Index].Input, _ = json.Marshal(currentInput)
+				}
+			}
+		}
+
+		// Handle finish
+		if choice.FinishReason == "tool_calls" {
+			// Parse accumulated tool arguments
+			for i := range toolCalls {
+				var tempInput map[string]string
+				json.Unmarshal(toolCalls[i].Input, &tempInput)
+				if argsStr, ok := tempInput["_args"]; ok && argsStr != "" {
+					var args map[string]interface{}
+					if err := json.Unmarshal([]byte(argsStr), &args); err == nil {
+						toolCalls[i].Input, _ = json.Marshal(args)
+					}
+				}
+				ch <- llm.StreamEvent{
+					Type:    "tool_use_end",
+					ToolUse: &toolCalls[i],
+				}
+			}
+		} else if choice.FinishReason == "stop" || choice.FinishReason == "length" {
+			ch <- llm.StreamEvent{Type: "done"}
+		}
+	}
+
+	// If no usage received, estimate tokens client-side
+	if !usageReceived {
+		inputTokens := estimateInputTokens(messages)
+		outputTokens := estimateTokens(textBuffer.String())
+		logger.Debug("DeepSeek: no usage from API, estimating: in=%d out=%d", inputTokens, outputTokens)
+		ch <- llm.StreamEvent{
+			Type:   "usage",
+			Input:  inputTokens,
+			Output: outputTokens,
+		}
+	}
+
+	if err := scanner.Err(); err != nil {
+		ch <- llm.StreamEvent{
+			Type: "error",
+			Err:  fmt.Errorf("stream read error: %w", err),
+		}
+	}
+}
+
+// estimateTokens estimates token count for text.
+// Uses similar logic to OpenAI provider.
+func estimateTokens(text string) int {
+	if text == "" {
+		return 0
+	}
+
+	cjkCount := 0
+	totalRunes := 0
+
+	for _, r := range text {
+		totalRunes++
+		// Check if character is CJK
+		if (r >= 0x4E00 && r <= 0x9FFF) || // CJK Unified Ideographs
+			(r >= 0x3400 && r <= 0x4DBF) || // CJK Extension A
+			(r >= 0x20000 && r <= 0x2A6DF) || // CJK Extension B
+			(r >= 0x2A700 && r <= 0x2B73F) || // CJK Extension C
+			(r >= 0x2B740 && r <= 0x2B81F) || // CJK Extension D
+			(r >= 0x2B820 && r <= 0x2CEAF) || // CJK Extension E
+			(r >= 0x3000 && r <= 0x303F) || // CJK Symbols
+			(r >= 0xFF00 && r <= 0xFFEF) { // Fullwidth Forms
+			cjkCount++
+		}
+	}
+
+	cjkRatio := float64(cjkCount) / float64(totalRunes)
+
+	if cjkRatio > 0.5 {
+		// Mostly CJK: ~1.5 chars per token
+		return int(float64(totalRunes) / 1.5)
+	} else {
+		// Mostly English: ~4 chars per token
+		return len(text) / 4
+	}
+}
+
+// estimateInputTokens estimates input token count.
+func estimateInputTokens(messages []session.Message) int {
+	total := 0
+	for _, msg := range messages {
+		// Add 4 tokens for message structure
+		total += 4
+
+		for _, content := range msg.Content {
+			if content.Type == "text" {
+				total += estimateTokens(content.Text)
+			} else if content.Type == "tool_use" {
+				// Estimate tool_use as JSON
+				data, _ := json.Marshal(content)
+				total += estimateTokens(string(data))
+			} else if content.Type == "tool_result" {
+				total += estimateTokens(content.Text)
+			}
+		}
+	}
+	return total
+}
+
+// convertMessages converts session messages to DeepSeek API format (OpenAI-compatible).
+func convertMessages(messages []session.Message) []map[string]interface{} {
+	var result []map[string]interface{}
+
+	for _, msg := range messages {
+		// Handle different message types
+		if msg.Role == "user" {
+			// Check if this is a tool result message
+			hasToolResult := false
+			for _, c := range msg.Content {
+				if c.Type == "tool_result" {
+					hasToolResult = true
+					break
+				}
+			}
+
+			if hasToolResult {
+				// Convert tool results to OpenAI format (role: "tool")
+				for _, c := range msg.Content {
+					if c.Type == "tool_result" {
+						result = append(result, map[string]interface{}{
+							"role":         "tool",
+							"tool_call_id": c.ToolUseID,
+							"content":      c.Text,
+						})
+					}
+				}
+			} else {
+				// Regular user message
+				var textParts []string
+				for _, c := range msg.Content {
+					if c.Type == "text" {
+						textParts = append(textParts, c.Text)
+					}
+				}
+				if len(textParts) > 0 {
+					result = append(result, map[string]interface{}{
+						"role":    "user",
+						"content": strings.Join(textParts, "\n"),
+					})
+				}
+			}
+		} else if msg.Role == "assistant" {
+			// Check if this message has tool calls
+			var toolCalls []map[string]interface{}
+			var textContent string
+
+			for _, c := range msg.Content {
+				if c.Type == "text" {
+					textContent = c.Text
+				} else if c.Type == "tool_use" {
+					// Convert to OpenAI tool_calls format
+					argsJSON, _ := json.Marshal(c.Input)
+					toolCalls = append(toolCalls, map[string]interface{}{
+						"id":   c.ID,
+						"type": "function",
+						"function": map[string]interface{}{
+							"name":      c.Name,
+							"arguments": string(argsJSON),
+						},
+					})
+				}
+			}
+
+			m := map[string]interface{}{
+				"role": "assistant",
+			}
+
+			if textContent != "" {
+				m["content"] = textContent
+			} else {
+				m["content"] = "" // OpenAI requires content field
+			}
+
+			if len(toolCalls) > 0 {
+				m["tool_calls"] = toolCalls
+			}
+
+			result = append(result, m)
+		} else {
+			// System or other roles - simple text
+			var textParts []string
+			for _, c := range msg.Content {
+				if c.Type == "text" {
+					textParts = append(textParts, c.Text)
+				}
+			}
+			if len(textParts) > 0 {
+				result = append(result, map[string]interface{}{
+					"role":    msg.Role,
+					"content": strings.Join(textParts, "\n"),
+				})
+			}
+		}
+	}
+
+	return result
+}
+
+// convertTools converts session tools to DeepSeek API format.
+func convertTools(tools []llm.ToolSchema) []map[string]interface{} {
+	var result []map[string]interface{}
+
+	for _, tool := range tools {
+		result = append(result, map[string]interface{}{
+			"type": "function",
+			"function": map[string]interface{}{
+				"name":        tool.Name,
+				"description": tool.Description,
+				"parameters":  tool.InputSchema,
+			},
+		})
+	}
+
+	return result
+}
+
+// Name returns the provider name.
+func (p *Provider) Name() string {
+	return "deepseek"
+}
+
+// CurrentModel returns the current model name.
+func (p *Provider) CurrentModel() string {
+	return p.model
+}
diff --git a/internal/llm/interface.go b/internal/llm/interface.go
index 0261414..f135def 100644
--- a/internal/llm/interface.go
+++ b/internal/llm/interface.go
@@ -14,7 +14,7 @@ type ToolSchema struct {
 
 // StreamEvent is emitted by Provider.Stream for each piece of the response.
 type StreamEvent struct {
-	Type    string // text_delta | tool_use_start | tool_use_delta | tool_use_end | usage | done | error
+	Type    string // text_delta | tool_use_start | tool_use_delta | tool_use_end | thinking_delta | thinking_done | usage | done | error
 	Delta   string
 	ToolUse *ToolUseBlock
 	Input   int // input token count (on usage event)
diff --git a/internal/logger/logger.go b/internal/logger/logger.go
index f174a2d..8bfa8be 100644
--- a/internal/logger/logger.go
+++ b/internal/logger/logger.go
@@ -33,31 +33,31 @@ func Init(verboseMode bool) {
 func Info(format string, args ...any) {
 	msg := fmt.Sprintf(format, args...)
 	if fileLogger != nil {
-		fileLogger.Printf("[INFO]  " + msg)
+		fileLogger.Print("[INFO]  " + msg)
 	}
 }
 
 func Debug(format string, args ...any) {
 	msg := fmt.Sprintf(format, args...)
 	if fileLogger != nil {
-		fileLogger.Printf("[DEBUG] " + msg)
+		fileLogger.Print("[DEBUG] " + msg)
 	}
 	if verbose {
-		fmt.Fprintf(os.Stderr, "\033[90m[debug] "+msg+"\033[0m\n")
+		fmt.Fprintf(os.Stderr, "\033[90m[debug] %s\033[0m\n", msg)
 	}
 }
 
 func Error(format string, args ...any) {
 	msg := fmt.Sprintf(format, args...)
 	if fileLogger != nil {
-		fileLogger.Printf("[ERROR] " + msg)
+		fileLogger.Print("[ERROR] " + msg)
 	}
-	fmt.Fprintf(os.Stderr, "\033[31m[error] "+msg+"\033[0m\n")
+	fmt.Fprintf(os.Stderr, "\033[31m[error] %s\033[0m\n", msg)
 }
 
 func Warn(format string, args ...any) {
 	msg := fmt.Sprintf(format, args...)
 	if fileLogger != nil {
-		fileLogger.Printf("[WARN]  " + msg)
+		fileLogger.Print("[WARN]  " + msg)
 	}
 }
diff --git a/internal/session/session.go b/internal/session/session.go
index 399ab0f..0546ec6 100644
--- a/internal/session/session.go
+++ b/internal/session/session.go
@@ -79,6 +79,11 @@ type Session struct {
 	Snapshots []FileSnapshot
 	Usage     TokenUsage
 	Model     string
+
+	// PendingSkill / PendingPrompt are set by /skill <name> <prompt>
+	// and consumed by the interactive loop to trigger RunWithSkill.
+	PendingSkillName string
+	PendingPrompt string
 }
 
 // New creates a fresh session.
diff --git a/internal/skills/builtin/apidoc.md b/internal/skills/builtin/apidoc.md
new file mode 100644
index 0000000..b0c76f6
--- /dev/null
+++ b/internal/skills/builtin/apidoc.md
@@ -0,0 +1,88 @@
+---
+name: apidoc
+aliases: [API文档, 接口文档, REST文档, OpenAPI, Swagger, api doc]
+description: 生成 REST API 接口文档（OpenAPI 3.0 格式）
+triggers:
+  - API文档
+  - 接口文档
+  - openapi
+  - swagger
+  - REST.*文档
+  - 写.*接口文档
+output_file: openapi.yaml
+---
+
+# Skill: API 接口文档
+
+你现在是一名 API 设计专家，请生成规范的 API 文档。
+
+## 输出格式
+
+同时生成两份文档：
+
+### 1. OpenAPI 3.0 YAML（机器可读）
+
+```yaml
+openapi: 3.0.3
+info:
+  title: API 名称
+  version: 1.0.0
+  description: |
+    API 说明
+  contact:
+    name: 团队名
+paths:
+  /resource:
+    get:
+      summary: 简短描述
+      operationId: uniqueId
+      tags: [分组]
+      parameters: []
+      responses:
+        '200':
+          description: 成功
+          content:
+            application/json:
+              schema:
+                $ref: '#/components/schemas/Resource'
+              example: {}
+        '400':
+          $ref: '#/components/responses/BadRequest'
+        '401':
+          $ref: '#/components/responses/Unauthorized'
+components:
+  schemas: {}
+  responses: {}
+  securitySchemes: {}
+```
+
+### 2. Markdown 可读版
+
+每个接口包含：
+- **接口名称与描述**
+- **HTTP 方法 + 路径**
+- **认证要求**
+- **请求参数表**（参数名 | 位置 | 类型 | 必须 | 说明 | 示例）
+- **请求体示例**（JSON）
+- **响应码说明表**（状态码 | 含义 | 示例）
+- **成功响应示例**
+- **错误响应示例**
+- **cURL 调用示例**
+
+## 设计规范
+
+- 路径使用小写 kebab-case，如 `/user-profiles`
+- 资源使用复数名词，如 `/users` 而非 `/user`
+- 分页参数统一：`page`、`page_size`、`total`
+- 错误响应统一格式：`{"error": {"code": "...", "message": "...", "details": []}}`
+- 时间字段使用 ISO 8601：`2024-01-01T00:00:00Z`
+- ID 字段说明类型（UUID/自增整数/ULID）
+
+## 操作方式
+
+1. 先读取源码（路由定义、handler、model 文件）
+2. 提取所有端点和数据结构
+3. 生成 OpenAPI YAML 文件
+4. 生成人可读的 Markdown 文档
+5. 校验 YAML 语法正确性
+6. 可调用tool function write_file
diff --git a/internal/skills/builtin/arch.md b/internal/skills/builtin/arch.md
new file mode 100644
index 0000000..fb31f45
--- /dev/null
+++ b/internal/skills/builtin/arch.md
@@ -0,0 +1,79 @@
+---
+name: arch
+aliases: [架构设计, 系统设计, 架构文档, architecture, system design]
+description: 编写技术架构设计文档（Architecture Design Document）
+triggers:
+  - 架构设计
+  - 系统设计
+  - architecture design
+  - 技术方案
+  - 设计文档
+  - arch doc
+output_file: arch.md
+---
+
+# Skill: 技术架构设计文档
+
+你现在是一名资深软件架构师，请按照以下结构生成一份完整的架构设计文档。
+
+## 必须包含的章节
+
+1. **架构概览**
+   - 系统目标与设计原则（如：高可用、可扩展、安全优先）
+   - 架构风格（微服务/单体/Serverless/事件驱动等）
+   - 关键技术决策及其理由（ADR 格式）
+
+2. **系统上下文（Context Diagram）**
+   - 系统边界：哪些在范围内，哪些是外部系统
+   - 主要用户类型和外部依赖
+   - 用 ASCII 图或 Mermaid 图表示
+
+3. **核心模块划分**
+   - 模块列表：名称、职责、对外接口
+   - 模块间依赖关系（避免循环依赖）
+   - 完整目录结构（精确到文件级）
+
+4. **数据架构**
+   - 数据模型（核心实体及关系）
+   - 存储选型（关系型/NoSQL/缓存/消息队列）及理由
+   - 数据流向（CRUD 操作如何流经系统）
+
+5. **接口设计**
+   - 对外 API 规范（REST/gRPC/GraphQL）
+   - 内部服务间通信协议
+   - 关键接口的请求/响应示例
+
+6. **关键技术选型**
+   - 技术栈表格（组件 | 选型 | 版本 | 选型理由 | 备选方案）
+
+7. **部署架构**
+   - 部署环境（云/私有/混合）
+   - 容器化/编排方案
+   - 网络拓扑、负载均衡、CDN
+
+8. **安全架构**
+   - 认证与授权模型
+   - 数据加密（传输/存储）
+   - 敏感操作审计
+
+9. **可观测性**
+   - 日志规范（结构化日志字段定义）
+   - 指标（核心 SLI/SLO）
+   - 链路追踪方案
+
+10. **演进路线**
+    - v1 → v2 的关键架构变更点
+    - 已知技术债及偿还计划
+
+## 图表规范
+
+- 所有架构图优先使用 Mermaid（可在 GitHub/GitLab 直接渲染）
+- 图表必须有标题和简短说明
+- 每个模块用方框，外部系统用圆角方框
+
+## 输出要求
+
+- 可调用tool function write_file
+- 完整 Markdown，含所有 Mermaid 图
+- 对每个关键决策说明"为什么选它，为什么不选备选方案"
+- 末尾附：术语表、参考资料
diff --git a/internal/skills/builtin/codedoc.md b/internal/skills/builtin/codedoc.md
new file mode 100644
index 0000000..5cdf47e
--- /dev/null
+++ b/internal/skills/builtin/codedoc.md
@@ -0,0 +1,82 @@
+---
+name: codedoc
+aliases: [代码文档, 代码说明, 代码注释, code documentation, code doc]
+description: 为代码生成完整的技术文档和注释
+triggers:
+  - 代码(文档|说明|注释)
+  - 给.*写.*注释
+  - 写.*文档.*代码
+  - code documentation
+  - document.*code
+  - API.*文档
+output_file: ""
+---
+
+# Skill: 代码文档生成
+
+你现在是一名技术文档专家，请为给定的代码生成完整、准确的文档。
+
+## 文档覆盖范围
+
+### 1. 文件级文档
+- 文件职责（一句话概括）
+- 包/模块说明
+- 关键依赖说明
+
+### 2. 函数/方法文档（每个公开函数必须有）
+```
+// FunctionName 简短描述（动词开头）。
+//
+// 详细描述（可选，超过一行时使用）。
+//
+// Parameters:
+//   - param1: 含义、取值范围、是否可为空
+//   - param2: 含义
+//
+// Returns:
+//   - 返回值含义
+//   - 错误情况
+//
+// Example:
+//   result, err := FunctionName(arg1, arg2)
+//
+// Note: 特殊行为、副作用、并发安全性
+```
+
+### 3. 类型/结构体文档
+- 类型用途
+- 每个字段的含义、单位、约束
+- 零值/默认值语义
+
+### 4. 常量/枚举文档
+- 每个值的含义
+- 使用场景
+
+### 5. 算法说明（复杂逻辑必须有）
+- 算法名称和时间/空间复杂度
+- 核心步骤编号说明
+- 边界条件处理
+
+## 注释风格规范
+
+按语言自动选择：
+- **Go**: GoDoc 格式
+- **Python**: Google Style Docstring
+- **TypeScript/JavaScript**: JSDoc
+- **Java/Kotlin**: JavaDoc
+- **Rust**: `///` rustdoc
+- **C/C++**: Doxygen
+
+## 额外输出（根据文件复杂度决定）
+
+- **README 段落**：该模块的使用示例
+- **流程图**：复杂业务逻辑的 Mermaid 流程图
+- **接口说明表**：公开 API 汇总表
+
+## 操作方式
+
+1. 先用 `read_file` 读取目标文件
+2. 分析代码结构，识别所有公开符号
+3. 为每个符号生成文档注释
+4. 用 `edit_file` 将注释写入原文件
+5. 输出文档变更摘要
diff --git a/internal/skills/builtin/debug.md b/internal/skills/builtin/debug.md
new file mode 100644
index 0000000..a6a730a
--- /dev/null
+++ b/internal/skills/builtin/debug.md
@@ -0,0 +1,93 @@
+---
+name: debug
+aliases: [调试, debug, 排查问题, 找bug, troubleshoot]
+description: 系统性问题排查与调试
+triggers:
+  - 调试
+  - debug
+  - 排查.*问题
+  - 找.*bug
+  - 为什么.*报错
+  - troubleshoot
+  - 报错了
+  - 有个bug
+  - 有.*bug
+output_file: ""
+---
+
+# Skill: 系统性调试
+
+你现在是一名调试专家，请系统性地分析和定位问题。
+
+## 调试方法论（5 Whys + 二分法）
+
+### Step 1: 信息收集（不做假设）
+- 完整的错误信息/堆栈跟踪
+- 复现步骤（能稳定复现 vs 偶发）
+- 环境信息（OS、语言版本、依赖版本）
+- 最近的变更（什么时候开始出现？）
+- 用 `search_files` 搜索相关错误关键词
+
+### Step 2: 假设生成
+列出所有可能的根因（脑暴，不过滤），按可能性排序
+
+### Step 3: 逐一验证（从最可能开始）
+- 用 `run_command` 执行诊断命令
+- 添加临时日志（最小改动）
+- 二分法缩小问题范围
+
+### Step 4: 根因确认
+- 确认能稳定复现
+- 确认修复后问题消失
+- 确认没有引入新问题
+
+### Step 5: 修复与预防
+- 最小化修复（只改必要的）
+- 添加回归测试（防止复现）
+- 记录根因和修复方式
+
+## 常用诊断命令模板
+
+```bash
+# 查看进程状态
+ps aux | grep <process>
+
+# 查看端口占用
+lsof -i :<port>
+netstat -tlnp | grep <port>
+
+# 查看系统日志
+journalctl -u <service> -n 100 --no-pager
+
+# 检查磁盘/内存
+df -h && free -h
+
+# Go 程序调试
+go run -race .          # 竞态检测
+GODEBUG=gctrace=1 ./app # GC 追踪
+dlv debug .             # 交互式调试器
+```
+
+## 调试输出格式
+
+```
+## 问题诊断报告
+
+**问题描述**: ...
+**复现率**: 必现/偶发(X%)
+**影响范围**: ...
+
+### 根因分析
+1. 现象: ...
+2. 第一层原因: ...（Why?）
+3. 第二层原因: ...（Why?）
+4. 根因: ...
+
+### 修复方案
+**立即修复**: [最小改动，最快恢复]
+**根本修复**: [彻底解决]
+**预防措施**: [防止复发]
+
+### 回归测试
+[验证修复的测试步骤]
+```
diff --git a/internal/skills/builtin/devplan.md b/internal/skills/builtin/devplan.md
new file mode 100644
index 0000000..af77777
--- /dev/null
+++ b/internal/skills/builtin/devplan.md
@@ -0,0 +1,80 @@
+---
+name: devplan
+aliases: [开发计划, 开发排期, 迭代计划, sprint plan, development plan, todo]
+description: 生成详细的开发计划与任务分解文档
+triggers:
+  - 开发计划
+  - 开发排期
+  - 迭代计划
+  - sprint
+  - 任务分解
+  - todo.*计划
+  - development plan
+output_file: todo.md
+---
+
+# Skill: 开发计划文档
+
+你现在是一名技术项目经理，请按照以下结构生成一份详细的开发计划文档。
+
+## 必须包含的章节
+
+1. **项目概况**
+   - 目标版本（如 v1.0 MVP）
+   - 总工期与团队规模
+   - 关键里程碑（Milestone）日期表
+
+2. **阶段划分（Phase）**
+   每个 Phase 包含：
+   - 阶段名称与目标
+   - 工期（周数）
+   - **完成标准**（DoD: Definition of Done，可验收的具体标准）
+
+3. **任务清单（细粒度）**
+   每个任务包含：
+   - [ ] 任务描述（动词开头，如"实现 xxx"、"编写 xxx"）
+   - 所属模块
+   - 优先级（P0/P1/P2）
+   - 预估工时
+   - 负责人（如知道）
+   - 依赖任务
+
+4. **技术风险登记表**
+   | 风险 | 可能性 | 影响 | 缓解措施 | 触发条件 |
+   |---|---|---|---|---|
+
+5. **依赖与阻塞项**
+   - 外部依赖（第三方 API、硬件、其他团队）
+   - 已知阻塞项及解决负责人
+
+6. **测试计划**
+   - 单元测试覆盖率目标
+   - 集成测试场景列表
+   - 性能测试基准
+
+7. **发布检查清单（Release Checklist）**
+   - [ ] 所有 P0 测试通过
+   - [ ] 文档更新完毕
+   - [ ] 安全扫描通过
+   - [ ] 性能基准达标
+   - …
+
+## 任务状态图例
+
+```
+⬜ 未开始  🔵 进行中  ✅ 已完成  🔶 阻塞中  ❌ 取消
+```
+
+## 工时估算规范
+
+- 最小粒度：0.5 天（4h）
+- 每个任务不超过 3 天，超过则继续拆分
+- 预留 20% buffer 用于联调和意外
+
+## 输出要求
+
+- 完整 Markdown，任务使用 `- [ ]` checkbox 格式
+- 按 Phase 组织，每 Phase 有完成标准
+- 末尾附风险登记表和发布清单
+- 如工期信息不足，给出合理估算并标注假设条件
+- 可调用tool function write_file
diff --git a/internal/skills/builtin/prd.md b/internal/skills/builtin/prd.md
new file mode 100644
index 0000000..bb6a934
--- /dev/null
+++ b/internal/skills/builtin/prd.md
@@ -0,0 +1,47 @@
+---
+name: prd
+aliases: [需求文档, 产品需求, 需求说明, product requirements, requirements doc]
+description: 编写专业的产品需求说明文档（PRD）
+triggers:
+  - 写.*需求(文档|说明)
+  - 产品需求
+  - PRD
+  - product requirement
+  - 需求文档
+output_file: prd.md
+---
+
+# Skill: 产品需求文档（PRD）
+
+你现在是一名资深产品经理，请按照以下结构生成一份专业的 PRD 文档。
+
+## 必须包含的章节
+
+1. **产品概述** — 背景、目标用户、产品定位（一句话定位）
+2. **问题陈述** — 用户痛点、现有方案的不足
+3. **核心功能需求** — 按优先级（P0/P1/P2）列出，每条需求含：
+   - 用户故事（As a... I want... So that...）
+   - 验收标准（Acceptance Criteria）
+   - 排除范围（Out of Scope）
+4. **非功能性需求** — 性能、安全、可用性、兼容性
+5. **用户体验设计要点** — 关键流程、UI 规范
+6. **数据需求** — 需要收集/存储/分析的数据
+7. **技术约束** — 已知的技术限制和依赖
+8. **成功指标（KPI）** — 可量化的成功标准
+9. **里程碑计划** — 分阶段交付计划（MVP → GA）
+10. **风险与假设** — 主要风险及缓解措施
+
+## 写作规范
+
+- 使用第三人称，避免"我觉得"等主观表述
+- 每条需求必须可测试、可验收
+- 优先级标注：P0=必须有，P1=重要，P2=锦上添花
+- 数据尽量量化（如"响应时间 < 200ms"而非"响应速度快"）
+- 输出为标准 Markdown，可直接保存为 `prd.md`
+
+## 输出要求
+
+- 可调用tool function write_file
+- 生成完整的 Markdown 文档
+- 文档末尾注明版本、作者（如已知）、日期
+- 如果信息不足，在对应章节标注 `[待补充]` 并列出需要确认的问题
diff --git a/internal/skills/builtin/refactor.md b/internal/skills/builtin/refactor.md
new file mode 100644
index 0000000..280d46a
--- /dev/null
+++ b/internal/skills/builtin/refactor.md
@@ -0,0 +1,63 @@
+---
+name: refactor
+aliases: [重构, 代码重构, refactor, clean code, 代码优化]
+description: 系统性代码重构：识别坏味道、制定方案、安全执行
+triggers:
+  - 重构
+  - refactor
+  - 代码优化
+  - clean.*code
+  - 改善代码
+output_file: ""
+---
+
+# Skill: 代码重构
+
+你现在是一名代码质量专家，请系统性地分析并重构代码。
+
+## 重构流程（必须严格遵守顺序）
+
+### Phase 1: 分析（不修改任何文件）
+1. 用 `read_file` 和 `search_files` 全面了解代码结构
+2. 识别代码坏味道（Bad Smells）：
+   - 过长函数（> 30 行）
+   - 过深嵌套（> 3 层）
+   - 重复代码（DRY 违反）
+   - 神秘数字（Magic Numbers）
+   - 过长参数列表（> 4 个参数）
+   - 数据泥团（Data Clumps）
+   - 过度耦合（Feature Envy）
+   - 注释掉的死代码
+3. 输出分析报告（不修改代码）
+
+### Phase 2: 制定方案
+列出具体重构操作，每条包含：
+- 重构手法名称（如：Extract Method、Rename Variable）
+- 影响范围（哪些文件/函数）
+- 预期收益
+- 风险评估
+- **必须：先有测试再重构（如无测试，先补测试）**
+
+### Phase 3: 执行（最小步骤，每步可验证）
+1. 每次只做一个独立的重构操作
+2. 修改后立即运行测试（`run_command "go test ./..."`）
+3. 测试失败则立即回滚（`/undo`）
+4. 测试通过才进行下一个重构
+
+### Phase 4: 验收
+- 运行全量测试，确保无回归
+- 对比重构前后的代码复杂度指标
+- 更新受影响的文档和注释
+
+## 重构原则
+
+- **小步快跑**：每次改动最小化，保持可回滚
+- **测试保障**：重构不改变外部行为，测试是保障
+- **命名优先**：好名字比好注释更重要
+- **SOLID 原则**：单一职责、开闭、里氏替换、接口隔离、依赖倒置
+
+## 禁止行为
+
+- ❌ 重构同时修复 Bug（分开做）
+- ❌ 重构同时添加新功能（分开做）
+- ❌ 在没有测试的情况下做大规模重构
diff --git a/internal/skills/builtin/review.md b/internal/skills/builtin/review.md
new file mode 100644
index 0000000..409fb90
--- /dev/null
+++ b/internal/skills/builtin/review.md
@@ -0,0 +1,81 @@
+---
+name: review
+aliases: [代码审查, code review, 审查代码, PR review]
+description: 对代码或 PR diff 进行专业的代码审查
+triggers:
+  - 代码审查
+  - code review
+  - review.*代码
+  - 审查.*代码
+  - PR review
+output_file: ""
+---
+
+# Skill: 代码审查（Code Review）
+
+你现在是一名资深工程师，请对代码进行全面、专业、建设性的审查。
+
+## 审查维度（按优先级）
+
+### 🔴 P0 — 必须修复（阻塞合并）
+- **正确性**：逻辑错误、边界条件遗漏、并发问题
+- **安全性**：SQL 注入、XSS、权限漏洞、敏感信息硬编码
+- **数据完整性**：缺少事务、竞态条件、数据丢失风险
+
+### 🟡 P1 — 强烈建议修复
+- **性能**：N+1 查询、不必要的全表扫描、内存泄漏
+- **错误处理**：未处理的错误、panic 风险、不完整的回滚
+- **测试**：缺少关键测试、测试覆盖率不足
+
+### 🔵 P2 — 建议优化
+- **可读性**：命名不清、注释缺失、函数过长
+- **可维护性**：重复代码、硬编码魔法值、过度耦合
+- **一致性**：与现有代码风格不符
+
+### ⚪ 建议（可选）
+- 更优雅的实现方式
+- 可以复用的已有工具/库
+
+## 审查报告格式
+
+```
+## 代码审查报告
+
+**文件**: xxx.go
+**审查人**: AI Assistant
+**审查时间**: YYYY-MM-DD
+
+### 总结
+[一段话总体评价，正面肯定 + 关键问题]
+
+### 问题列表
+
+#### [P0] 并发安全问题 (line 42)
+**问题**: `map` 在多 goroutine 下读写未加锁
+**风险**: 运行时 panic
+**建议**: 
+```go
+// 修改前
+m[key] = value
+
+// 修改后
+mu.Lock()
+m[key] = value
+mu.Unlock()
+```
+
+[按 P0 → P1 → P2 顺序列出所有问题]
+
+### 亮点
+[值得表扬的好代码]
+
+### 建议
+[整体性改进建议]
+```
+
+## 操作方式
+
+1. 读取所有相关文件（或分析 `git diff` 输出）
+2. 系统性分析每个审查维度
+3. 生成结构化审查报告
+4. 对 P0 问题，可直接用 `edit_file` 提供修复建议（不自动应用，征求用户同意）
diff --git a/internal/skills/builtin/testplan.md b/internal/skills/builtin/testplan.md
new file mode 100644
index 0000000..c292ab4
--- /dev/null
+++ b/internal/skills/builtin/testplan.md
@@ -0,0 +1,76 @@
+---
+name: testplan
+aliases: [测试计划, 测试方案, 测试用例, test plan, test cases]
+description: 生成测试计划、测试用例和测试报告模板
+triggers:
+  - 测试计划
+  - 测试方案
+  - 测试用例
+  - test plan
+  - 写测试
+  - 生成测试
+output_file: testplan.md
+---
+
+# Skill: 测试计划与用例
+
+你现在是一名资深测试工程师，请生成全面的测试文档。
+
+## 测试策略金字塔
+
+```
+        [E2E 测试]      ← 少量，覆盖核心用户路径
+       [集成测试]        ← 适量，覆盖模块间交互
+    [单元测试]            ← 大量，覆盖所有业务逻辑
+```
+
+## 必须包含的测试类型
+
+### 1. 单元测试用例
+每个测试用例包含：
+- 测试目标（函数/方法名）
+- 前置条件（Arrange）
+- 执行步骤（Act）
+- 断言（Assert）
+- 边界值：空值、最大值、最小值、非法输入
+- 异常路径：错误处理、超时、网络异常
+
+### 2. 集成测试场景
+- 核心业务流程的完整路径测试
+- 数据库/缓存/消息队列的读写验证
+- 第三方服务的 Mock 测试
+
+### 3. 性能测试基准
+| 指标 | 目标值 | 测试工具 |
+|---|---|---|
+| P50 响应时间 | < Xms | wrk/k6 |
+| P99 响应时间 | < Xms | |
+| QPS | > X | |
+| 错误率 | < 0.1% | |
+
+### 4. 安全测试清单
+- [ ] SQL 注入
+- [ ] XSS
+- [ ] CSRF
+- [ ] 权限越权
+- [ ] 敏感信息泄露
+- [ ] 速率限制
+
+## 测试用例格式
+
+```
+TC-001: [模块名] 正常场景描述
+前置条件: xxx
+输入: xxx
+预期输出: xxx
+优先级: P0
+自动化: 是/否
+```
+
+## 操作方式
+
+1. 读取源码，分析所有公开函数和业务逻辑
+2. 识别关键路径和边界条件
+3. 生成对应语言的测试代码框架（Go/pytest/Jest 等）
+4. 用 `write_file` 写入测试文件
+5. 运行已有测试套件验证（`run_command`）
diff --git a/internal/skills/skill.go b/internal/skills/skill.go
new file mode 100644
index 0000000..6bbe685
--- /dev/null
+++ b/internal/skills/skill.go
@@ -0,0 +1,276 @@
+// Package skills implements the aicoder skill system.
+// Skills are Markdown files with a YAML front-matter header that define
+// specialized AI personas and structured output guidance. They are embedded
+// into the binary at compile time and can also be loaded from
+// ~/.aicoder/skills/ for user-defined skills.
+package skills
+
+import (
+	"bufio"
+	"bytes"
+	"embed"
+	"fmt"
+	"os"
+	"path/filepath"
+	"regexp"
+	"strings"
+)
+
+//go:embed builtin/*.md
+var builtinFS embed.FS
+
+// Skill represents one loaded skill definition.
+type Skill struct {
+	// Metadata (from YAML front-matter)
+	Name        string   // canonical name, e.g. "prd"
+	Aliases     []string // alternative trigger phrases
+	Description string   // one-line description
+	Triggers    []string // regex patterns for auto-detection
+	OutputFile  string   // suggested output filename (empty = no file)
+
+	// Content (everything after the front-matter)
+	Prompt string // full skill prompt injected as additional context
+
+	// Compiled matchers
+	compiled []*regexp.Regexp
+}
+
+// Matches returns true if the input string triggers this skill.
+func (s *Skill) Matches(input string) bool {
+	lower := strings.ToLower(input)
+	// Check explicit name / aliases
+	for _, alias := range append([]string{s.Name}, s.Aliases...) {
+		if strings.Contains(lower, strings.ToLower(alias)) {
+			return true
+		}
+	}
+	// Check regex triggers
+	for _, re := range s.compiled {
+		if re.MatchString(input) {
+			return true
+		}
+	}
+	return false
+}
+
+// ─── Registry ─────────────────────────────────────────────────────────────────
+
+// Registry holds all loaded skills.
+type Registry struct {
+	skills []*Skill
+	byName map[string]*Skill
+}
+
+// Global is the default registry, populated at startup.
+var Global = &Registry{byName: map[string]*Skill{}}
+
+// Load loads all built-in skills plus any user skills from ~/.aicoder/skills/.
+func Load() error {
+	if err := loadBuiltins(); err != nil {
+		return fmt.Errorf("load built-in skills: %w", err)
+	}
+	loadUserSkills() // best-effort; errors are silently ignored
+	return nil
+}
+
+func loadBuiltins() error {
+	entries, err := builtinFS.ReadDir("builtin")
+	if err != nil {
+		return err
+	}
+	for _, e := range entries {
+		if e.IsDir() || !strings.HasSuffix(e.Name(), ".md") {
+			continue
+		}
+		data, err := builtinFS.ReadFile("builtin/" + e.Name())
+		if err != nil {
+			return err
+		}
+		skill, err := parseSkill(data)
+		if err != nil {
+			return fmt.Errorf("parse %s: %w", e.Name(), err)
+		}
+		Global.register(skill)
+	}
+	return nil
+}
+
+func loadUserSkills() {
+	home, err := os.UserHomeDir()
+	if err != nil {
+		return
+	}
+	dir := filepath.Join(home, ".aicoder", "skills")
+	entries, err := os.ReadDir(dir)
+	if err != nil {
+		return // directory doesn't exist yet — normal
+	}
+	for _, e := range entries {
+		if e.IsDir() || !strings.HasSuffix(e.Name(), ".md") {
+			continue
+		}
+		data, err := os.ReadFile(filepath.Join(dir, e.Name()))
+		if err != nil {
+			continue
+		}
+		skill, err := parseSkill(data)
+		if err != nil {
+			continue
+		}
+		skill.Name = "user:" + skill.Name // namespace user skills
+		Global.register(skill)
+	}
+}
+
+func (r *Registry) register(s *Skill) {
+	r.skills = append(r.skills, s)
+	r.byName[s.Name] = s
+	// Also index by aliases
+	for _, alias := range s.Aliases {
+		key := strings.ToLower(alias)
+		if _, exists := r.byName[key]; !exists {
+			r.byName[key] = s
+		}
+	}
+}
+
+// All returns all registered skills.
+func (r *Registry) All() []*Skill { return r.skills }
+
+// Get returns a skill by name (case-insensitive), or nil.
+func (r *Registry) Get(name string) *Skill {
+	if s, ok := r.byName[strings.ToLower(name)]; ok {
+		return s
+	}
+	// Partial match on name
+	lower := strings.ToLower(name)
+	for _, s := range r.skills {
+		if strings.HasPrefix(s.Name, lower) {
+			return s
+		}
+	}
+	return nil
+}
+
+// Detect finds the best-matching skill for a user input string.
+// Returns nil if no skill matches.
+func (r *Registry) Detect(input string) *Skill {
+	for _, s := range r.skills {
+		if s.Matches(input) {
+			return s
+		}
+	}
+	return nil
+}
+
+// ─── Parser ───────────────────────────────────────────────────────────────────
+
+// parseSkill parses a Markdown file with YAML-like front-matter.
+// Front-matter is delimited by --- lines.
+func parseSkill(data []byte) (*Skill, error) {
+	scanner := bufio.NewScanner(bytes.NewReader(data))
+	s := &Skill{}
+	inFrontMatter := false
+	frontMatterDone := false
+	var bodyLines []string
+	lineNum := 0
+
+	for scanner.Scan() {
+		line := scanner.Text()
+		lineNum++
+
+		if lineNum == 1 && line == "---" {
+			inFrontMatter = true
+			continue
+		}
+		if inFrontMatter && line == "---" {
+			inFrontMatter = false
+			frontMatterDone = true
+			continue
+		}
+		if inFrontMatter {
+			parseFrontMatterLine(s, line)
+			continue
+		}
+		if frontMatterDone {
+			bodyLines = append(bodyLines, line)
+		}
+	}
+
+	if s.Name == "" {
+		return nil, fmt.Errorf("skill is missing 'name' field in front-matter")
+	}
+
+	s.Prompt = strings.Join(bodyLines, "\n")
+
+	// Compile regex triggers
+	for _, pattern := range s.Triggers {
+		re, err := regexp.Compile("(?i)" + pattern)
+		if err == nil {
+			s.compiled = append(s.compiled, re)
+		}
+	}
+
+	return s, nil
+}
+
+// parseFrontMatterLine handles one key: value line from the front-matter.
+func parseFrontMatterLine(s *Skill, line string) {
+	// Handle list items: "  - value"
+	trimmed := strings.TrimSpace(line)
+
+	// Key: value
+	if idx := strings.Index(trimmed, ":"); idx > 0 {
+		key := strings.TrimSpace(trimmed[:idx])
+		val := strings.TrimSpace(trimmed[idx+1:])
+
+		switch key {
+		case "name":
+			s.Name = val
+		case "description":
+			s.Description = val
+		case "output_file":
+			s.OutputFile = strings.Trim(val, `"`)
+		case "aliases", "triggers":
+			// Inline list: [a, b, c]
+			if strings.HasPrefix(val, "[") && strings.HasSuffix(val, "]") {
+				items := splitList(val[1 : len(val)-1])
+				switch key {
+				case "aliases":
+					s.Aliases = items
+				case "triggers":
+					s.Triggers = items
+				}
+			}
+			// Multi-line list handled by list item case below
+		}
+		return
+	}
+
+	// List item: "  - value"
+	if strings.HasPrefix(trimmed, "- ") {
+		// We need context from the previous key — track with a simple heuristic:
+		// if the skill has more triggers than aliases, we're in triggers; else aliases
+		val := strings.TrimSpace(strings.TrimPrefix(trimmed, "- "))
+		val = strings.Trim(val, `"'`)
+		// Assign to whichever list was last being populated
+		// Heuristic: triggers tend to have regex chars
+		if strings.ContainsAny(val, ".*+?()[]") || len(s.Triggers) < len(s.Aliases) {
+			s.Triggers = append(s.Triggers, val)
+		} else {
+			s.Aliases = append(s.Aliases, val)
+		}
+	}
+}
+
+func splitList(s string) []string {
+	var result []string
+	for _, part := range strings.Split(s, ",") {
+		v := strings.TrimSpace(part)
+		v = strings.Trim(v, `"'`)
+		if v != "" {
+			result = append(result, v)
+		}
+	}
+	return result
+}
diff --git a/internal/skills/skill_integration_test.go b/internal/skills/skill_integration_test.go
new file mode 100644
index 0000000..0f71617
--- /dev/null
+++ b/internal/skills/skill_integration_test.go
@@ -0,0 +1,199 @@
+package skills
+
+import (
+	"strings"
+	"testing"
+)
+
+// TestAllBuiltinTriggersDetected verifies every built-in skill can be detected
+// by at least one of its documented trigger phrases.
+func TestAllBuiltinTriggersDetected(t *testing.T) {
+	if err := Load(); err != nil {
+		t.Fatal(err)
+	}
+
+	cases := []struct {
+		phrase   string
+		wantName string
+	}{
+		// prd
+		{"帮我写产品需求文档", "prd"},
+		{"请生成一份PRD", "prd"},
+		{"product requirement document", "prd"},
+		{"需求说明书怎么写", "prd"},
+		// arch
+		{"帮我做架构设计", "arch"},
+		{"请写一份系统设计文档", "arch"},
+		{"architecture design for microservice", "arch"},
+		{"技术方案怎么写", "arch"},
+		// devplan
+		{"制定开发计划", "devplan"},
+		{"帮我做开发排期", "devplan"},
+		{"sprint planning", "devplan"},
+		{"任务分解一下", "devplan"},
+		// codedoc
+		{"给这个文件写代码注释", "codedoc"},
+		{"帮我写代码文档", "codedoc"},
+		{"code documentation for this module", "codedoc"},
+		// apidoc
+		{"生成API文档", "apidoc"},
+		{"写一份接口文档", "apidoc"},
+		{"openapi spec", "apidoc"},
+		{"swagger documentation", "apidoc"},
+		// testplan
+		{"写测试计划", "testplan"},
+		{"生成测试用例", "testplan"},
+		{"test plan for this feature", "testplan"},
+		// refactor
+		{"帮我重构这段代码", "refactor"},
+		{"refactor this function", "refactor"},
+		{"代码优化一下", "refactor"},
+		// review
+		{"code review一下", "review"},
+		{"帮我做代码审查", "review"},
+		{"review this PR", "review"},
+		// debug
+		{"帮我debug这个问题", "debug"},
+		{"程序报错了怎么排查", "debug"},
+		{"troubleshoot this error", "debug"},
+		{"这里有个bug", "debug"},
+	}
+
+	for _, c := range cases {
+		t.Run(c.phrase, func(t *testing.T) {
+			s := Global.Detect(c.phrase)
+			if s == nil {
+				t.Errorf("Detect(%q) = nil, want %q", c.phrase, c.wantName)
+				return
+			}
+			if s.Name != c.wantName {
+				t.Errorf("Detect(%q) = %q, want %q", c.phrase, s.Name, c.wantName)
+			}
+		})
+	}
+}
+
+// TestSkillPromptsNotEmpty verifies every skill has a meaningful prompt.
+func TestSkillPromptsNotEmpty(t *testing.T) {
+	if err := Load(); err != nil {
+		t.Fatal(err)
+	}
+	for _, s := range Global.All() {
+		if len(strings.TrimSpace(s.Prompt)) < 100 {
+			t.Errorf("skill %q prompt too short (%d chars)", s.Name, len(s.Prompt))
+		}
+	}
+}
+
+// TestSkillOutputFiles verifies output_file values are sane.
+func TestSkillOutputFiles(t *testing.T) {
+	if err := Load(); err != nil {
+		t.Fatal(err)
+	}
+	withFile := map[string]string{
+		"prd":     "prd.md",
+		"arch":    "arch.md",
+		"devplan": "todo.md",
+		"apidoc":  "openapi.yaml",
+		"testplan": "testplan.md",
+	}
+	for name, want := range withFile {
+		s := Global.Get(name)
+		if s == nil {
+			t.Errorf("skill %q not found", name)
+			continue
+		}
+		if s.OutputFile != want {
+			t.Errorf("skill %q: output_file = %q, want %q", name, s.OutputFile, want)
+		}
+	}
+	// These skills edit files in-place, no output_file
+	noFile := []string{"codedoc", "refactor", "review", "debug"}
+	for _, name := range noFile {
+		s := Global.Get(name)
+		if s == nil {
+			t.Errorf("skill %q not found", name)
+			continue
+		}
+		if s.OutputFile != "" {
+			t.Errorf("skill %q should have empty output_file, got %q", name, s.OutputFile)
+		}
+	}
+}
+
+// TestNoFalsePositives verifies common non-skill inputs don't match any skill.
+func TestNoFalsePositives(t *testing.T) {
+	if err := Load(); err != nil {
+		t.Fatal(err)
+	}
+	inputs := []string{
+		"hello",
+		"今天天气怎么样",
+		"what time is it",
+		"ls -la",
+		"帮我看看这段代码",
+		"解释一下这个函数",
+		"git status",
+		"how does TCP work",
+	}
+	for _, input := range inputs {
+		s := Global.Detect(input)
+		if s != nil {
+			t.Errorf("Detect(%q) = %q, want no match (false positive)", input, s.Name)
+		}
+	}
+}
+
+// TestRegistryAll verifies All() returns all 9 built-in skills.
+func TestRegistryAll(t *testing.T) {
+	// Reset and reload
+	Global = &Registry{byName: map[string]*Skill{}}
+	if err := Load(); err != nil {
+		t.Fatal(err)
+	}
+	all := Global.All()
+	if len(all) < 9 {
+		t.Errorf("expected at least 9 built-in skills, got %d", len(all))
+	}
+}
+
+// TestUserSkillCreation simulates user skill file parsing.
+func TestUserSkillCreation(t *testing.T) {
+	raw := []byte(`---
+name: mycompany-pr
+aliases: [PR模板, pull request]
+description: 生成符合公司规范的 PR 描述
+triggers:
+  - PR模板
+  - pull request.*描述
+output_file: ""
+---
+
+# 公司 PR 规范
+
+## PR 标题格式
+[类型] 简短描述
+
+## 必填章节
+- **背景**: 为什么要做这个改动
+- **方案**: 怎么做的
+- **测试**: 如何验证
+- **风险**: 可能影响什么
+`)
+	s, err := parseSkill(raw)
+	if err != nil {
+		t.Fatalf("parseSkill failed: %v", err)
+	}
+	if s.Name != "mycompany-pr" {
+		t.Errorf("unexpected name: %s", s.Name)
+	}
+	if len(s.Aliases) < 2 {
+		t.Errorf("expected 2 aliases, got %d", len(s.Aliases))
+	}
+	if !strings.Contains(s.Prompt, "PR 规范") {
+		t.Error("prompt should contain skill content")
+	}
+	if s.OutputFile != "" {
+		t.Errorf("expected empty output_file, got %q", s.OutputFile)
+	}
+}
diff --git a/internal/skills/skill_test.go b/internal/skills/skill_test.go
new file mode 100644
index 0000000..5ff5ac3
--- /dev/null
+++ b/internal/skills/skill_test.go
@@ -0,0 +1,156 @@
+package skills
+
+import (
+	"strings"
+	"testing"
+)
+
+func TestLoadBuiltins(t *testing.T) {
+	r := &Registry{byName: map[string]*Skill{}}
+	if err := loadBuiltins(); err != nil {
+		t.Fatalf("loadBuiltins failed: %v", err)
+	}
+	// Verify Global registry has been populated
+	if err := Load(); err != nil {
+		t.Fatal(err)
+	}
+	all := Global.All()
+	if len(all) == 0 {
+		t.Fatal("expected at least one built-in skill")
+	}
+	_ = r
+}
+
+func TestBuiltinSkillNames(t *testing.T) {
+	_ = Load()
+	expectedNames := []string{"prd", "arch", "devplan", "codedoc", "apidoc", "testplan", "refactor", "review", "debug"}
+	for _, name := range expectedNames {
+		s := Global.Get(name)
+		if s == nil {
+			t.Errorf("expected skill %q to be registered", name)
+			continue
+		}
+		if s.Description == "" {
+			t.Errorf("skill %q has empty description", name)
+		}
+		if s.Prompt == "" {
+			t.Errorf("skill %q has empty prompt", name)
+		}
+	}
+}
+
+func TestSkillDetect(t *testing.T) {
+	_ = Load()
+	cases := []struct {
+		input    string
+		expected string
+	}{
+		{"帮我写一个产品需求文档", "prd"},
+		{"请生成架构设计文档", "arch"},
+		{"制定开发计划", "devplan"},
+		{"给这段代码写文档注释", "codedoc"},
+		{"生成API文档", "apidoc"},
+		{"写测试计划", "testplan"},
+		{"帮我重构这个函数", "refactor"},
+		{"review一下这段代码", "review"},
+		{"这个程序报错了，帮我debug", "debug"},
+	}
+	for _, c := range cases {
+		s := Global.Detect(c.input)
+		if s == nil {
+			t.Errorf("Detect(%q) = nil, want %q", c.input, c.expected)
+			continue
+		}
+		if s.Name != c.expected {
+			t.Errorf("Detect(%q) = %q, want %q", c.input, s.Name, c.expected)
+		}
+	}
+}
+
+func TestSkillGet(t *testing.T) {
+	_ = Load()
+	s := Global.Get("prd")
+	if s == nil {
+		t.Fatal("expected to find prd skill")
+	}
+	if s.Name != "prd" {
+		t.Errorf("unexpected name: %s", s.Name)
+	}
+}
+
+func TestSkillGetCaseInsensitive(t *testing.T) {
+	_ = Load()
+	if Global.Get("PRD") == nil {
+		t.Error("expected case-insensitive Get to work")
+	}
+}
+
+func TestSkillGetPartialMatch(t *testing.T) {
+	_ = Load()
+	// "dev" should match "devplan"
+	s := Global.Get("dev")
+	if s == nil {
+		t.Error("expected partial match on 'dev'")
+	}
+}
+
+func TestParseSkill(t *testing.T) {
+	raw := []byte(`---
+name: myskill
+aliases: [alias1, alias2]
+description: A test skill
+triggers:
+  - test.*skill
+  - myskill
+output_file: out.md
+---
+
+# My Skill Content
+
+This is the skill prompt.
+`)
+	s, err := parseSkill(raw)
+	if err != nil {
+		t.Fatalf("parseSkill failed: %v", err)
+	}
+	if s.Name != "myskill" {
+		t.Errorf("unexpected name: %s", s.Name)
+	}
+	if s.Description != "A test skill" {
+		t.Errorf("unexpected description: %s", s.Description)
+	}
+	if s.OutputFile != "out.md" {
+		t.Errorf("unexpected output_file: %s", s.OutputFile)
+	}
+	if !strings.Contains(s.Prompt, "My Skill Content") {
+		t.Errorf("prompt missing content: %s", s.Prompt)
+	}
+	if len(s.Aliases) < 2 {
+		t.Errorf("expected 2 aliases, got %d: %v", len(s.Aliases), s.Aliases)
+	}
+}
+
+func TestSkillMatches(t *testing.T) {
+	s := &Skill{
+		Name:    "test",
+		Aliases: []string{"测试技能"},
+	}
+	if !s.Matches("这是一个测试技能") {
+		t.Error("expected alias to match")
+	}
+	if !s.Matches("test skill") {
+		t.Error("expected name to match")
+	}
+	if s.Matches("completely unrelated") {
+		t.Error("expected no match")
+	}
+}
+
+func TestDetectNoMatch(t *testing.T) {
+	_ = Load()
+	// Something completely unrelated
+	s := Global.Detect("the weather is nice today 今天天气很好")
+	if s != nil {
+		t.Errorf("expected no match, got skill %q", s.Name)
+	}
+}
diff --git a/internal/slash/commands.go b/internal/slash/commands.go
index 7bae8f7..7ee374b 100644
--- a/internal/slash/commands.go
+++ b/internal/slash/commands.go
@@ -10,6 +10,7 @@ import (
 
 	"github.com/iminders/aicoder/internal/config"
 	"github.com/iminders/aicoder/internal/session"
+	"github.com/iminders/aicoder/internal/skills"
 	"github.com/iminders/aicoder/internal/ui"
 	"github.com/iminders/aicoder/pkg/diff"
 	"github.com/iminders/aicoder/pkg/version"
@@ -17,12 +18,32 @@ import (
 
 // Handler processes a slash command string. Returns true if the program should exit.
 type Handler struct {
-	sess *session.Session
-	cfg  *config.Config
+	sess    *session.Session
+	cfg     *config.Config
+	printer func(...interface{}) // For printing output (can be tea.Program.Println or fmt.Println)
 }
 
 func NewHandler(sess *session.Session, cfg *config.Config) *Handler {
-	return &Handler{sess: sess, cfg: cfg}
+	return &Handler{
+		sess:    sess,
+		cfg:     cfg,
+		printer: func(args ...interface{}) { fmt.Println(args...) }, // Default to fmt.Println
+	}
+}
+
+// SetPrinter sets the print function (use tea.Program.Println for TUI mode)
+func (h *Handler) SetPrinter(printer func(...interface{})) {
+	h.printer = printer
+}
+
+// println is a helper that uses the configured printer
+func (h *Handler) println(args ...interface{}) {
+	h.printer(args...)
+}
+
+// printf is a helper for formatted printing
+func (h *Handler) printf(format string, args ...interface{}) {
+	h.printer(fmt.Sprintf(format, args...))
 }
 
 // Handle dispatches a slash command. Returns (handled, shouldExit).
@@ -63,6 +84,8 @@ func (h *Handler) Handle(input string) (handled bool, shouldExit bool) {
 		h.cmdSave()
 	case "/tools":
 		h.cmdTools()
+	case "/skill", "/skills":
+		return h.cmdSkill(args)
 	default:
 		ui.PrintWarn(fmt.Sprintf("未知命令: %s  (输入 /help 查看所有命令)", cmd))
 	}
@@ -83,13 +106,17 @@ func (h *Handler) cmdHelp() {
 │ /cost         │ 查看 Token 用量和费用估算                         │
 │ /model [m]    │ 查看或切换 AI 模型                               │
 │ /config       │ 查看当前配置                                      │
-│ /init         │ 在当前目录初始化 AICODER.md                       │
+│ /init         │ 在当前目录初始化 .AICODER.md                       │
 │ /sessions     │ 列出历史会话                                      │
 │ /save         │ 手动保存当前会话                                  │
 │ /tools        │ 列出所有可用工具                                  │
+│ /skill list          │ 列出所有内置 Skill                             │
+│ /skill <名称>        │ 显示 Skill 详情                               │
+│ /skill <名称> <提示> │ 以指定 Skill 模式运行                         │
+│ /skill new <名称>    │ 创建自定义 Skill 模板                         │
 │ /exit         │ 退出程序                                          │
 └───────────────┴──────────────────────────────────────────────────┘`
-	fmt.Println(help)
+	h.printer(help)
 }
 
 func (h *Handler) cmdClear() {
@@ -103,7 +130,7 @@ func (h *Handler) cmdHistory() {
 		ui.PrintInfo("暂无对话历史")
 		return
 	}
-	fmt.Printf("\033[1m对话历史 (%d 条消息):\033[0m\n", len(msgs))
+	h.printf("\033[1m对话历史 (%d 条消息):\033[0m\n", len(msgs))
 	ui.PrintDivider()
 	for i, m := range msgs {
 		if m.Role == "system" {
@@ -130,7 +157,7 @@ func (h *Handler) cmdHistory() {
 				break
 			}
 		}
-		fmt.Printf("%s[%d] %s %s\033[0m\n", color, i, icon, preview)
+		h.printf("%s[%d] %s %s\033[0m\n", color, i, icon, preview)
 	}
 	ui.PrintDivider()
 }
@@ -154,13 +181,13 @@ func (h *Handler) cmdDiff() {
 		ui.PrintInfo("本次会话暂无文件变更")
 		return
 	}
-	fmt.Printf("\033[1m本次会话文件变更 (%d 个文件):\033[0m\n", len(changes))
+	h.printf("\033[1m本次会话文件变更 (%d 个文件):\033[0m\n", len(changes))
 	ui.PrintDivider()
 	for path, after := range changes {
 		before, _ := os.ReadFile(path)
 		d := diff.ColorDiff(string(before), string(after), path)
 		if d != "" {
-			fmt.Print(d)
+			h.printer(d)
 		}
 	}
 }
@@ -186,7 +213,7 @@ func (h *Handler) cmdCommit(args []string) {
 		return
 	}
 	ui.PrintSuccess("已提交: " + msg)
-	fmt.Println(string(out))
+	h.println(string(out))
 }
 
 func (h *Handler) cmdCost() {
@@ -194,25 +221,25 @@ func (h *Handler) cmdCost() {
 	model := h.sess.Model
 	est := usage.CostEstimate(model)
 	ui.PrintDivider()
-	fmt.Printf("  \033[1m模型:\033[0m      %s\n", model)
-	fmt.Printf("  \033[1m输入 tokens:\033[0m %d\n", usage.InputTokens)
-	fmt.Printf("  \033[1m输出 tokens:\033[0m %d\n", usage.OutputTokens)
-	fmt.Printf("  \033[1m费用估算:\033[0m   $%.4f USD\n", est)
+	h.printf("  \033[1m模型:\033[0m      %s\n", model)
+	h.printf("  \033[1m输入 tokens:\033[0m %d\n", usage.InputTokens)
+	h.printf("  \033[1m输出 tokens:\033[0m %d\n", usage.OutputTokens)
+	h.printf("  \033[1m费用估算:\033[0m   $%.4f USD\n", est)
 	ui.PrintDivider()
 }
 
 func (h *Handler) cmdModel(args []string) {
 	if len(args) == 0 {
-		fmt.Printf("当前模型: \033[1m%s\033[0m\n", h.sess.Model)
-		fmt.Println("可用模型示例:")
+		h.printf("当前模型: \033[1m%s\033[0m\n", h.sess.Model)
+		h.println("可用模型示例:")
 		models := []string{
 			"claude-opus-4-5", "claude-sonnet-4-5", "claude-haiku-4-5-20251001",
 			"gpt-4o", "gpt-4o-mini",
 		}
 		for _, m := range models {
-			fmt.Printf("  - %s\n", m)
+			h.printf("  - %s\n", m)
 		}
-		fmt.Println("用法: /model <model-name>")
+		h.println("用法: /model <model-name>")
 		return
 	}
 	newModel := args[0]
@@ -223,26 +250,26 @@ func (h *Handler) cmdModel(args []string) {
 
 func (h *Handler) cmdConfig(args []string) {
 	_ = args
-	fmt.Printf("\033[1m当前配置:\033[0m\n")
+	h.printf("\033[1m当前配置:\033[0m\n")
 	ui.PrintDivider()
-	fmt.Printf("  provider:          %s\n", h.cfg.Provider)
-	fmt.Printf("  model:             %s\n", h.cfg.Model)
-	fmt.Printf("  maxTokens:         %d\n", h.cfg.MaxTokens)
-	fmt.Printf("  autoApprove:       %v\n", h.cfg.AutoApprove)
-	fmt.Printf("  autoApproveReads:  %v\n", h.cfg.AutoApproveReads)
-	fmt.Printf("  backupOnWrite:     %v\n", h.cfg.BackupOnWrite)
-	fmt.Printf("  theme:             %s\n", h.cfg.Theme)
-	fmt.Printf("  language:          %s\n", h.cfg.Language)
+	h.printf("  provider:          %s\n", h.cfg.Provider)
+	h.printf("  model:             %s\n", h.cfg.Model)
+	h.printf("  maxTokens:         %d\n", h.cfg.MaxTokens)
+	h.printf("  autoApprove:       %v\n", h.cfg.AutoApprove)
+	h.printf("  autoApproveReads:  %v\n", h.cfg.AutoApproveReads)
+	h.printf("  backupOnWrite:     %v\n", h.cfg.BackupOnWrite)
+	h.printf("  theme:             %s\n", h.cfg.Theme)
+	h.printf("  language:          %s\n", h.cfg.Language)
 	if h.cfg.Proxy != "" {
-		fmt.Printf("  proxy:             %s\n", h.cfg.Proxy)
+		h.printf("  proxy:             %s\n", h.cfg.Proxy)
 	}
 	ui.PrintDivider()
 }
 
 func (h *Handler) cmdInit() {
-	path := filepath.Join("AICODER.md")
+	path := filepath.Join(".AICODER.md")
 	if _, err := os.Stat(path); err == nil {
-		ui.PrintWarn("AICODER.md 已存在，跳过初始化")
+		ui.PrintWarn(".AICODER.md 已存在，跳过初始化")
 		return
 	}
 	template := fmt.Sprintf(`# 项目说明
@@ -257,14 +284,19 @@ func (h *Handler) cmdInit() {
 # 注意事项
 <!-- 描述需要特别注意的事项，例如禁止修改的文件、特殊依赖等 -->
 
+# 工具使用规范
+<!-- 说明如何使用 aicoder 的工具，例如文件操作、命令执行等 -->
+允许使用web_search工具进行联网搜索
+允许git clone 到third_party目录, 但禁止直接修改第三方代码
+
 _由 aicoder v%s 生成于 %s_
 `, version.Version, time.Now().Format("2006-01-02"))
 
 	if err := os.WriteFile(path, []byte(template), 0644); err != nil {
-		ui.PrintError("创建 AICODER.md 失败: " + err.Error())
+		ui.PrintError("创建 .AICODER.md 失败: " + err.Error())
 		return
 	}
-	ui.PrintSuccess("已创建 AICODER.md，请编辑它来描述您的项目")
+	ui.PrintSuccess("已创建 .AICODER.md，请编辑它来描述您的项目")
 }
 func (h *Handler) cmdSessions() {
 	home, err := os.UserHomeDir()
@@ -279,7 +311,7 @@ func (h *Handler) cmdSessions() {
 		return
 	}
 
-	fmt.Printf("\033[1m历史会话 (%d 个):\033[0m\n", len(entries))
+	h.printf("\033[1m历史会话 (%d 个):\033[0m\n", len(entries))
 	ui.PrintDivider()
 
 	// Show most recent 20, newest first
@@ -300,10 +332,10 @@ func (h *Handler) cmdSessions() {
 		if info != nil {
 			modTime = info.ModTime().Format("2006-01-02 15:04")
 		}
-		fmt.Printf("  \033[36m%s\033[0m  %s  %s\n", name[:min(len(name), 20)], modTime, size)
+		h.printf("  \033[36m%s\033[0m  %s  %s\n", name[:min(len(name), 20)], modTime, size)
 	}
 	ui.PrintDivider()
-	fmt.Println("  提示：会话文件保存在", dir)
+	h.println("  提示：会话文件保存在", dir)
 }
 
 func (h *Handler) cmdSave() {
@@ -317,7 +349,7 @@ func (h *Handler) cmdSave() {
 func (h *Handler) cmdTools() {
 	// Import tools package to list all registered tools
 	// We use a type assertion via the session's known tool names
-	fmt.Printf("\033[1m已注册工具:\033[0m\n")
+	h.printf("\033[1m已注册工具:\033[0m\n")
 	ui.PrintDivider()
 
 	// Tool metadata is stored in the global registry; we query it via the session
@@ -332,19 +364,174 @@ func (h *Handler) cmdTools() {
 		{"run_command",     "中", "执行 Shell 命令"},
 		{"run_background",  "中", "后台启动长时进程"},
 		{"grep_search",     "低", "全目录正则搜索"},
-		{"web_search",      "低", "联网搜索（需 TAVILY_API_KEY）"},
+		{"web_search",      "低", "联网搜索"},
 	}
 	for _, r := range rows {
 		riskColor := "\033[32m"
 		if r.risk == "中" { riskColor = "\033[33m" }
 		if r.risk == "高" { riskColor = "\033[31m" }
-		fmt.Printf("  %-18s %s[%s]\033[0m  %s\n", r.name, riskColor, r.risk, r.desc)
+		h.printf("  %-18s %s[%s]\033[0m  %s\n", r.name, riskColor, r.risk, r.desc)
 	}
 	ui.PrintDivider()
-	fmt.Println("  MCP 工具以 <server>__<tool> 格式列出（连接后可见）")
+	h.println("  MCP 工具以 <server>__<tool> 格式列出（连接后可见）")
 }
 
 func min(a, b int) int {
 	if a < b { return a }
 	return b
 }
+
+
+
+// cmdSkill handles: /skill list | /skill <name> | /skill <name> <prompt> | /skill new <name>
+// It returns (handled bool, shouldExit bool) so the caller can optionally
+// hand off to the agent with a skill override.
+func (h *Handler) cmdSkill(args []string) (bool, bool) {
+
+	if len(args) == 0 || args[0] == "list" {
+		h.printSkillList()
+		return true, false
+	}
+
+	if args[0] == "new" {
+		if len(args) < 2 {
+			ui.PrintWarn("用法: /skill new <名称>")
+			return true, false
+		}
+		h.createUserSkill(args[1])
+		return true, false
+	}
+
+	// /skill <name> [optional prompt...]
+	skillName := args[0]
+	sk := skills.Global.Get(skillName)
+	if sk == nil {
+		ui.PrintError(fmt.Sprintf("未找到 Skill %q，输入 /skill list 查看所有可用 Skill", skillName))
+		return true, false
+	}
+
+	if len(args) == 1 {
+		// Show skill details
+		h.printSkillDetail(sk)
+		return true, false
+	}
+
+	// /skill <name> <user prompt> — signal caller to run agent with this skill
+	// We store the pending skill+prompt in session metadata and return a special
+	// sentinel so the interactive loop can handle it.
+	prompt := strings.Join(args[1:], " ")
+	fmt.Printf("\033[90m[Skill %q 已激活，正在处理: %s]\033[0m\n", sk.Name, prompt)
+	// Inject skill directly — store on session for the loop to pick up
+	h.sess.PendingSkillName = sk.Name
+	h.sess.PendingPrompt = prompt
+	return true, false
+}
+
+func (h *Handler) printSkillList() {
+	all := skills.Global.All()
+	fmt.Printf("\033[1m内置 Skill (%d 个):\033[0m\n", len(all))
+	ui.PrintDivider()
+	for _, s := range all {
+		tag := "\033[34m[内置]\033[0m"
+		if strings.HasPrefix(s.Name, "user:") {
+			tag = "\033[32m[自定义]\033[0m"
+		}
+		outFile := ""
+		if s.OutputFile != "" {
+			outFile = fmt.Sprintf("  \033[90m→ %s\033[0m", s.OutputFile)
+		}
+		fmt.Printf("  %-12s %s  %s%s\n", s.Name, tag, s.Description, outFile)
+	}
+	ui.PrintDivider()
+	fmt.Println("  用法: /skill <名称> <你的需求描述>")
+	fmt.Println("  示例: /skill prd 电商平台用户评价系统")
+	fmt.Println("  自动触发: 直接描述需求，aicoder 会自动匹配合适的 Skill")
+}
+
+func (h *Handler) printSkillDetail(sk *skills.Skill) {
+	fmt.Printf("\n\033[1m🎯 Skill: %s\033[0m\n", sk.Name)
+	ui.PrintDivider()
+	fmt.Printf("  \033[1m描述:\033[0m    %s\n", sk.Description)
+	if len(sk.Aliases) > 0 {
+		fmt.Printf("  \033[1m别名:\033[0m    %s\n", strings.Join(sk.Aliases, ", "))
+	}
+	if len(sk.Triggers) > 0 {
+		fmt.Printf("  \033[1m触发词:\033[0m  %s\n", strings.Join(sk.Triggers[:min(3, len(sk.Triggers))], " | "))
+	}
+	if sk.OutputFile != "" {
+		fmt.Printf("  \033[1m输出文件:\033[0m %s\n", sk.OutputFile)
+	}
+	ui.PrintDivider()
+	// Show first 10 lines of the prompt as preview
+	lines := strings.Split(sk.Prompt, "\n")
+	preview := lines
+	truncated := false
+	if len(lines) > 12 {
+		preview = lines[:12]
+		truncated = true
+	}
+	fmt.Println("\033[90m" + strings.Join(preview, "\n") + "\033[0m")
+	if truncated {
+		fmt.Printf("\033[90m... (共 %d 行) ...\033[0m\n", len(lines))
+	}
+	ui.PrintDivider()
+	fmt.Printf("  运行: /skill %s <你的需求描述>\n\n", sk.Name)
+}
+
+func (h *Handler) createUserSkill(name string) {
+	home, err := os.UserHomeDir()
+	if err != nil {
+		ui.PrintError("无法获取 home 目录: " + err.Error())
+		return
+	}
+	dir := filepath.Join(home, ".aicoder", "skills")
+	if err := os.MkdirAll(dir, 0700); err != nil {
+		ui.PrintError("创建目录失败: " + err.Error())
+		return
+	}
+	path := filepath.Join(dir, name+".md")
+	if _, err := os.Stat(path); err == nil {
+		ui.PrintWarn(fmt.Sprintf("Skill %q 已存在: %s", name, path))
+		return
+	}
+
+	template := fmt.Sprintf(`---
+name: %s
+aliases: []
+description: 在这里填写 Skill 的一句话描述
+triggers:
+  - 触发关键词1
+  - 触发关键词2
+output_file: output.md
+---
+
+# Skill: %s
+
+在这里描述这个 Skill 的职责和使用场景。
+
+## 输出结构
+
+1. **章节一** — 说明
+2. **章节二** — 说明
+
+## 写作规范
+
+- 规范1
+- 规范2
+
+## 操作方式
+
+1. 先用工具收集信息
+2. 按结构生成文档
+3. 保存到输出文件
+
+_由 aicoder v%s 生成于 %s_
+`, name, name, version.Version, time.Now().Format("2006-01-02"))
+
+	if err := os.WriteFile(path, []byte(template), 0644); err != nil {
+		ui.PrintError("创建 Skill 失败: " + err.Error())
+		return
+	}
+	ui.PrintSuccess(fmt.Sprintf("已创建自定义 Skill 模板: %s", path))
+	fmt.Println("  请编辑该文件，然后重启 aicoder 或输入 /skill list 刷新")
+}
diff --git a/internal/slash/completion.go b/internal/slash/completion.go
index fd56037..35008b7 100644
--- a/internal/slash/completion.go
+++ b/internal/slash/completion.go
@@ -21,7 +21,7 @@ func AllCommands() []CommandInfo {
 		{"/cost", "查看 Token 用量和费用估算", "/cost"},
 		{"/model", "查看或切换 AI 模型", "/model [name]"},
 		{"/config", "查看或修改配置", "/config [set key value]"},
-		{"/init", "在当前目录初始化 AICODER.md", "/init"},
+		{"/init", "在当前目录初始化 .AICODER.md", "/init"},
 		{"/sessions", "列出历史会话", "/sessions"},
 		{"/save", "保存当前会话到文件", "/save [filename]"},
 		{"/tools", "列出可用工具", "/tools"},
diff --git a/internal/tools/filesystem/tools.go b/internal/tools/filesystem/tools.go
index b701db4..b3c7213 100644
--- a/internal/tools/filesystem/tools.go
+++ b/internal/tools/filesystem/tools.go
@@ -72,19 +72,27 @@ func (t *WriteFileTool) Execute(_ context.Context, raw json.RawMessage) (*tools.
 	if err := json.Unmarshal(raw, &in); err != nil {
 		return &tools.Result{IsError: true, Content: err.Error()}, nil
 	}
+	if in.Path == "" {
+		return &tools.Result{IsError: true, Content: "path cannot be empty"}, nil
+	}
 	if err := checkSandbox(in.Path); err != nil {
 		return &tools.Result{IsError: true, Content: err.Error()}, nil
 	}
-	// Snapshot before state
-	var before []byte
-	before, _ = os.ReadFile(in.Path)
 
 	if err := os.MkdirAll(filepath.Dir(in.Path), 0755); err != nil {
 		return &tools.Result{IsError: true, Content: err.Error()}, nil
 	}
+
+	// Snapshot before state
+	var before []byte
+	if _, err := os.Stat(in.Path); err == nil {
+		before, _ = os.ReadFile(in.Path)
+	}
+
 	if err := os.WriteFile(in.Path, []byte(in.Content), 0644); err != nil {
 		return &tools.Result{IsError: true, Content: err.Error()}, nil
 	}
+	
 	if SnapshotFunc != nil {
 		SnapshotFunc("write_file", fmt.Sprintf("%d", time.Now().UnixNano()), in.Path, before, []byte(in.Content))
 	}
@@ -98,7 +106,7 @@ type EditFileTool struct{}
 func (t *EditFileTool) Name() string          { return "edit_file" }
 func (t *EditFileTool) Risk() tools.RiskLevel { return tools.RiskMedium }
 func (t *EditFileTool) Description() string {
-	return "Edit a file by replacing an exact old_string with new_string. old_string must match exactly once."
+	return "Edit a file by replacing an exact old_string with new_string if it exist. old_string must match exactly once."
 }
 func (t *EditFileTool) Schema() json.RawMessage {
 	return json.RawMessage(`{
diff --git a/internal/tools/search/web_search.go b/internal/tools/search/web_search.go
index 8be48f6..5fbd596 100644
--- a/internal/tools/search/web_search.go
+++ b/internal/tools/search/web_search.go
@@ -6,10 +6,8 @@ import (
 	"fmt"
 	"io"
 	"net/http"
-	"net/url"
 	"os"
 	"strings"
-	"time"
 
 	"github.com/iminders/aicoder/internal/tools"
 )
@@ -19,7 +17,7 @@ type WebSearchTool struct{}
 func (t *WebSearchTool) Name() string          { return "web_search" }
 func (t *WebSearchTool) Risk() tools.RiskLevel { return tools.RiskLow }
 func (t *WebSearchTool) Description() string {
-	return "Search the web using a search API. Requires SEARCH_API_KEY and SEARCH_ENGINE_ID environment variables to be set. Returns top search results with titles, snippets, and URLs."
+	return "Search the web using Tavily API. Returns top search results with titles, snippets, and URLs."
 }
 
 func (t *WebSearchTool) Schema() json.RawMessage {
@@ -46,12 +44,13 @@ type searchResult struct {
 	Snippet string `json:"snippet"`
 }
 
-type googleSearchResponse struct {
-	Items []struct {
+type tavilySearchResponse struct {
+	Results []struct {
 		Title   string `json:"title"`
-		Link    string `json:"link"`
-		Snippet string `json:"snippet"`
-	} `json:"items"`
+		URL     string `json:"url"`
+		Content string `json:"content"`
+		Score   float64 `json:"score"`
+	} `json:"results"`
 }
 
 func (t *WebSearchTool) Execute(ctx context.Context, raw json.RawMessage) (*tools.Result, error) {
@@ -72,24 +71,21 @@ func (t *WebSearchTool) Execute(ctx context.Context, raw json.RawMessage) (*tool
 	}
 
 	// 检查环境变量
-	apiKey := os.Getenv("SEARCH_API_KEY")
-	engineID := os.Getenv("SEARCH_ENGINE_ID")
+	apiKey := os.Getenv("TAVILY_API_KEY")
 
-	if apiKey == "" || engineID == "" {
+	if apiKey == "" {
 		return &tools.Result{
 			IsError: true,
-			Content: "Web search is not configured. Please set SEARCH_API_KEY and SEARCH_ENGINE_ID environment variables.\n\n" +
-				"To use Google Custom Search:\n" +
-				"1. Get API key from: https://developers.google.com/custom-search/v1/overview\n" +
-				"2. Create search engine at: https://programmablesearchengine.google.com/\n" +
-				"3. Set environment variables:\n" +
-				"   export SEARCH_API_KEY=\"your-api-key\"\n" +
-				"   export SEARCH_ENGINE_ID=\"your-engine-id\"",
+			Content: "Web search is not configured. Please set TAVILY_API_KEY environment variable.\n\n" +
+				"To use Tavily Search:\n" +
+				"1. Get API key from: https://tavily.com/\n" +
+				"2. Set environment variable:\n" +
+				"   export TAVILY_API_KEY=\"tvly-your-api-key\"",
 		}, nil
 	}
 
 	// 执行搜索
-	results, err := t.googleSearch(ctx, in.Query, in.NumResults, in.Language, apiKey, engineID)
+	results, err := t.tavilySearch(ctx, in.Query, in.NumResults, apiKey)
 	if err != nil {
 		return &tools.Result{IsError: true, Content: fmt.Sprintf("Search failed: %v", err)}, nil
 	}
@@ -117,29 +113,32 @@ func (t *WebSearchTool) Execute(ctx context.Context, raw json.RawMessage) (*tool
 	}, nil
 }
 
-// googleSearch 使用 Google Custom Search API 进行搜索
-func (t *WebSearchTool) googleSearch(ctx context.Context, query string, numResults int, language, apiKey, engineID string) ([]searchResult, error) {
-	// 构建 API URL
-	baseURL := "https://www.googleapis.com/customsearch/v1"
-	params := url.Values{}
-	params.Set("key", apiKey)
-	params.Set("cx", engineID)
-	params.Set("q", query)
-	params.Set("num", fmt.Sprintf("%d", numResults))
-	params.Set("lr", "lang_"+language)
+// tavilySearch 使用 Tavily Search API 进行搜索
+func (t *WebSearchTool) tavilySearch(ctx context.Context, query string, numResults int, apiKey string) ([]searchResult, error) {
+	// 构建请求体
+	requestBody := map[string]interface{}{
+		"api_key":        apiKey,
+		"query":          query,
+		"max_results":    numResults,
+		"search_depth":   "basic",
+		"include_answer": false,
+	}
 
-	apiURL := baseURL + "?" + params.Encode()
+	bodyJSON, err := json.Marshal(requestBody)
+	if err != nil {
+		return nil, fmt.Errorf("failed to marshal request: %w", err)
+	}
 
 	// 创建 HTTP 请求
-	req, err := http.NewRequestWithContext(ctx, "GET", apiURL, nil)
+	req, err := http.NewRequestWithContext(ctx, "POST", "https://api.tavily.com/search", strings.NewReader(string(bodyJSON)))
 	if err != nil {
 		return nil, fmt.Errorf("failed to create request: %w", err)
 	}
 
-	// 设置超时
-	client := &http.Client{
-		Timeout: 30 * time.Second,
-	}
+	req.Header.Set("Content-Type", "application/json")
+
+	// 使用不带超时的client，完全依赖context控制超时
+	client := &http.Client{}
 
 	// 发送请求
 	resp, err := client.Do(req)
@@ -155,18 +154,18 @@ func (t *WebSearchTool) googleSearch(ctx context.Context, query string, numResul
 	}
 
 	// 解析响应
-	var searchResp googleSearchResponse
+	var searchResp tavilySearchResponse
 	if err := json.NewDecoder(resp.Body).Decode(&searchResp); err != nil {
 		return nil, fmt.Errorf("failed to parse response: %w", err)
 	}
 
 	// 转换结果
-	results := make([]searchResult, 0, len(searchResp.Items))
-	for _, item := range searchResp.Items {
+	results := make([]searchResult, 0, len(searchResp.Results))
+	for _, item := range searchResp.Results {
 		results = append(results, searchResult{
 			Title:   item.Title,
-			Link:    item.Link,
-			Snippet: item.Snippet,
+			Link:    item.URL,
+			Snippet: item.Content,
 		})
 	}
 
diff --git a/internal/tools/search/web_search_test.go b/internal/tools/search/web_search_test.go
index 27baa49..782bf3f 100644
--- a/internal/tools/search/web_search_test.go
+++ b/internal/tools/search/web_search_test.go
@@ -10,8 +10,7 @@ import (
 
 func TestWebSearchTool_NoConfig(t *testing.T) {
 	// 确保环境变量未设置
-	os.Unsetenv("SEARCH_API_KEY")
-	os.Unsetenv("SEARCH_ENGINE_ID")
+	os.Unsetenv("TAVILY_API_KEY")
 
 	tool := &WebSearchTool{}
 
@@ -34,8 +33,8 @@ func TestWebSearchTool_NoConfig(t *testing.T) {
 		t.Errorf("Expected configuration error message, got: %s", result.Content)
 	}
 
-	if !strings.Contains(result.Content, "SEARCH_API_KEY") {
-		t.Errorf("Expected mention of SEARCH_API_KEY, got: %s", result.Content)
+	if !strings.Contains(result.Content, "TAVILY_API_KEY") {
+		t.Errorf("Expected mention of TAVILY_API_KEY, got: %s", result.Content)
 	}
 }
 
@@ -172,11 +171,10 @@ func TestWebSearchTool_ToolInterface(t *testing.T) {
 // TestWebSearchTool_Integration 是集成测试,需要真实的 API 凭证
 // 默认跳过,可以通过设置环境变量来运行
 func TestWebSearchTool_Integration(t *testing.T) {
-	apiKey := os.Getenv("SEARCH_API_KEY")
-	engineID := os.Getenv("SEARCH_ENGINE_ID")
+	apiKey := os.Getenv("TAVILY_API_KEY")
 
-	if apiKey == "" || engineID == "" {
-		t.Skip("Skipping integration test: SEARCH_API_KEY or SEARCH_ENGINE_ID not set")
+	if apiKey == "" {
+		t.Skip("Skipping integration test: TAVILY_API_KEY not set")
 	}
 
 	tool := &WebSearchTool{}
diff --git a/internal/ui/app.go b/internal/ui/app.go
index 617f213..e8b1fca 100644
--- a/internal/ui/app.go
+++ b/internal/ui/app.go
@@ -37,6 +37,7 @@ type Model struct {
 	currentOutput strings.Builder
 	inputBuffer   string
 	statusText    string
+	completions   []string // Current autocomplete suggestions
 
 	// Metadata
 	modelName string
@@ -44,8 +45,10 @@ type Model struct {
 	tokens    int
 
 	// Callbacks
-	onSubmit func(string) error
-	onCancel func()
+	onSubmit      func(string) error
+	onCancel      func()
+	onSlashCmd    func(string) (handled bool, shouldExit bool)
+	slashCommands []string // For autocomplete
 }
 
 // Message represents a chat message.
@@ -84,7 +87,8 @@ func (m *Model) Update(msg tea.Msg) (tea.Model, tea.Cmd) {
 		if m.markdown != nil {
 			m.markdown.SetWidth(msg.Width - 4)
 		}
-		return m, nil
+		// Clear screen on resize to avoid artifacts
+		return m, tea.ClearScreen
 
 	case tea.KeyMsg:
 		return m.handleKeyPress(msg)
@@ -121,10 +125,26 @@ func (m *Model) handleKeyPress(msg tea.KeyMsg) (tea.Model, tea.Cmd) {
 		m.quitting = true
 		return m, tea.Quit
 
-	case tea.KeyEnter:
+	case tea.KeyCtrlD:
+		// Ctrl+D submits the input
 		if m.state == StateIdle && m.inputBuffer != "" {
-			input := m.inputBuffer
+			input := strings.TrimSpace(m.inputBuffer)
 			m.inputBuffer = ""
+
+			// Check for slash commands
+			if strings.HasPrefix(input, "/") && m.onSlashCmd != nil {
+				// Execute slash command (will use tea.Println for output)
+				handled, shouldExit := m.onSlashCmd(input)
+				if shouldExit {
+					m.quitting = true
+					return m, tea.Quit
+				}
+				if handled {
+					// After slash command, just return to keep TUI running
+					return m, nil
+				}
+			}
+
 			m.messages = append(m.messages, Message{
 				Role:    "user",
 				Content: input,
@@ -138,15 +158,41 @@ func (m *Model) handleKeyPress(msg tea.KeyMsg) (tea.Model, tea.Cmd) {
 		}
 		return m, nil
 
+	case tea.KeyEnter:
+		// Enter adds a newline to support multi-line input
+		if m.state == StateIdle {
+			m.inputBuffer += "\n"
+		}
+		return m, nil
+
+	case tea.KeyTab:
+		// Tab for slash command autocomplete
+		if m.state == StateIdle && strings.HasPrefix(m.inputBuffer, "/") {
+			m.autocompleteSlashCommand()
+		}
+		return m, nil
+
 	case tea.KeyBackspace:
 		if len(m.inputBuffer) > 0 {
 			m.inputBuffer = m.inputBuffer[:len(m.inputBuffer)-1]
+			// Update completions if editing a slash command
+			if strings.HasPrefix(m.inputBuffer, "/") {
+				m.updateCompletions()
+			} else {
+				m.completions = nil
+			}
 		}
 		return m, nil
 
 	case tea.KeyRunes:
 		if m.state == StateIdle {
 			m.inputBuffer += string(msg.Runes)
+			// Update completions if typing a slash command
+			if strings.HasPrefix(m.inputBuffer, "/") {
+				m.updateCompletions()
+			} else {
+				m.completions = nil
+			}
 		}
 		return m, nil
 	}
@@ -162,38 +208,77 @@ func (m *Model) View() string {
 
 	var b strings.Builder
 
-	// Status bar
+	// Status bar (always at top)
 	b.WriteString(m.renderStatusBar())
 	b.WriteString("\n\n")
 
+	// Calculate available height for content
+	// Reserve space for: status bar (3 lines), input area (4 lines), margins
+	availableHeight := m.height - 7
+	if availableHeight < 5 {
+		availableHeight = 5
+	}
+
+	// Collect all content lines
+	var contentLines []string
+
 	// Messages
 	for _, msg := range m.messages {
-		b.WriteString(m.renderMessage(msg))
-		b.WriteString("\n")
+		rendered := m.renderMessage(msg)
+		lines := strings.Split(rendered, "\n")
+		contentLines = append(contentLines, lines...)
 	}
 
 	// Current streaming output
 	if m.currentOutput.Len() > 0 {
-		b.WriteString(m.theme.InfoStyle.Render("Assistant: "))
-		b.WriteString(m.currentOutput.String())
+		output := m.theme.InfoStyle.Render("Assistant: ") + m.currentOutput.String()
+		lines := strings.Split(output, "\n")
+		contentLines = append(contentLines, lines...)
+	}
+
+	// Show only the last N lines that fit in the available height
+	startLine := 0
+	if len(contentLines) > availableHeight {
+		startLine = len(contentLines) - availableHeight
+	}
+
+	for i := startLine; i < len(contentLines); i++ {
+		b.WriteString(contentLines[i])
 		b.WriteString("\n")
 	}
 
-	// Spinner for thinking/streaming state
+	// Spinner for thinking/streaming state (on same line)
 	if m.state == StateThinking {
-		b.WriteString(m.spinner.View())
-		b.WriteString(" Thinking...\n")
+		b.WriteString(m.spinner.View() + " Thinking...")
 	} else if m.state == StateStreaming {
-		b.WriteString(m.spinner.View())
-		b.WriteString(" Streaming...\n")
+		b.WriteString(m.spinner.View() + " Streaming...")
 	}
 
-	// Input prompt
+	// Input prompt (always at bottom)
 	if m.state == StateIdle {
 		b.WriteString("\n")
 		b.WriteString(m.theme.PromptStyle.Render("> "))
 		b.WriteString(m.inputBuffer)
 		b.WriteString(lipgloss.NewStyle().Foreground(m.theme.Muted).Render("_"))
+		b.WriteString("\n")
+
+		// Show completions if available
+		if len(m.completions) > 0 {
+			b.WriteString(m.theme.SubtitleStyle.Render("Completions: "))
+			for i, comp := range m.completions {
+				if i > 0 {
+					b.WriteString(", ")
+				}
+				b.WriteString(m.theme.InfoStyle.Render(comp))
+				if i >= 4 { // Limit to 5 suggestions
+					b.WriteString(fmt.Sprintf(" ... (%d more)", len(m.completions)-5))
+					break
+				}
+			}
+			b.WriteString("\n")
+		}
+
+		b.WriteString(m.theme.SubtitleStyle.Render("Press Ctrl+D to submit, Enter for new line, Tab for autocomplete"))
 	}
 
 	return b.String()
@@ -251,6 +336,81 @@ func (m *Model) SetOnCancel(fn func()) {
 	m.onCancel = fn
 }
 
+// SetOnSlashCommand sets the callback for slash commands.
+func (m *Model) SetOnSlashCommand(fn func(string) (bool, bool)) {
+	m.onSlashCmd = fn
+}
+
+// SetSlashCommands sets the list of available slash commands for autocomplete.
+func (m *Model) SetSlashCommands(commands []string) {
+	m.slashCommands = commands
+}
+
+// autocompleteSlashCommand attempts to autocomplete the current slash command.
+func (m *Model) autocompleteSlashCommand() {
+	if len(m.slashCommands) == 0 {
+		return
+	}
+
+	input := strings.TrimSpace(m.inputBuffer)
+	if !strings.HasPrefix(input, "/") {
+		return
+	}
+
+	// Find matching commands
+	var matches []string
+	for _, cmd := range m.slashCommands {
+		if strings.HasPrefix(cmd, input) {
+			matches = append(matches, cmd)
+		}
+	}
+
+	// If exactly one match, complete it
+	if len(matches) == 1 {
+		m.inputBuffer = matches[0] + " "
+		m.completions = nil
+	} else if len(matches) > 1 {
+		// Find common prefix
+		commonPrefix := matches[0]
+		for _, match := range matches[1:] {
+			for i := 0; i < len(commonPrefix) && i < len(match); i++ {
+				if commonPrefix[i] != match[i] {
+					commonPrefix = commonPrefix[:i]
+					break
+				}
+			}
+		}
+		if len(commonPrefix) > len(input) {
+			m.inputBuffer = commonPrefix
+		}
+		m.completions = matches
+	}
+}
+
+// updateCompletions updates the list of completion suggestions.
+func (m *Model) updateCompletions() {
+	if len(m.slashCommands) == 0 {
+		m.completions = nil
+		return
+	}
+
+	input := strings.TrimSpace(m.inputBuffer)
+	if !strings.HasPrefix(input, "/") {
+		m.completions = nil
+		return
+	}
+
+	// Find matching commands
+	var matches []string
+	for _, cmd := range m.slashCommands {
+		if strings.HasPrefix(cmd, input) {
+			matches = append(matches, cmd)
+		}
+	}
+
+	m.completions = matches
+}
+
 // AddMessage adds a message to the history.
 func (m *Model) AddMessage(role, content string) {
 	m.messages = append(m.messages, Message{
diff --git a/scripts/proxy_tool.py b/scripts/proxy_tool.py
new file mode 100644
index 0000000..8f37289
--- /dev/null
+++ b/scripts/proxy_tool.py
@@ -0,0 +1,65 @@
+from flask import Flask, request, Response
+import requests
+import json
+
+app = Flask(__name__)
+
+# 目标：你本地运行 DeepSeek R1 的真实端口
+TARGET_SERVER = "http://127.0.0.1:10002"
+
+@app.route('/v1/chat/completions', methods=['POST'])
+@app.route('/chat/completions', methods=['POST'])
+def proxy_deepseek():
+    req_data = request.get_json(silent=True)
+
+
+    print("="*120 + "\n")
+    print(json.dumps(req_data, indent=4, ensure_ascii=False))
+    
+    # --- 打印逻辑保持不变 ---
+    print(f"\n🚀 [Intercepted] Model: {req_data.get('model')}")
+
+    # 1. 清洗 Headers：移除会导致冲突的传输层 Header
+    excluded_headers = ['content-encoding', 'content-length', 'transfer-encoding', 'connection', 'host']
+    headers = {
+        k: v for k, v in request.headers 
+        if k.lower() not in excluded_headers
+    }
+
+    # 2. 转发请求
+    is_stream = req_data.get('stream', False) if req_data else False
+    
+    try:
+        resp = requests.post(
+            f"{TARGET_SERVER}{request.full_path}",
+            json=req_data,
+            headers=headers,
+            stream=is_stream,
+            timeout=300 # DeepSeek R1 推理较慢，建议增加超时
+        )
+
+        # 3. 构造响应并清洗返回的 Headers
+        def generate():
+            for chunk in resp.iter_content(chunk_size=None): # chunk_size=None 保持原始分块
+                yield chunk
+
+        # 同样移除返回时的冲突 Header
+        resp_headers = [
+            (k, v) for k, v in resp.raw.headers.items() 
+            if k.lower() not in excluded_headers
+        ]
+
+        return Response(
+            generate() if is_stream else resp.content,
+            status=resp.status_code,
+            headers=resp_headers
+        )
+    except Exception as e:
+        print(f"❌ Proxy Error: {e}")
+        return jsonify({"error": str(e)}), 500
+
+if __name__ == '__main__':
+    # 启动代理，监听 10004
+    print("Proxy is running on http://127.0.0.1:10004")
+    print(f"Forwarding to DeepSeek R1 at {TARGET_SERVER}")
+    app.run(port=10004, debug=False)