Merge pull request #34 from cexll/cce-worktree-master-20251129-111802-997076000

feat: add parallel execution support to codex-wrapper

.claude-plugin/marketplace.json

@@ -175,6 +175,87 @@
       "skills": [
         "./skills/SKILL.md"
       ]
+    },
+    {
+      "name": "codex-cli",
+      "source": "./skills/codex/",
+      "description": "Execute Codex CLI for code analysis, refactoring, and automated code changes with file references (@syntax) and structured output",
+      "version": "1.0.0",
+      "author": {
+        "name": "Claude Code Dev Workflows",
+        "url": "https://github.com/cexll/myclaude"
+      },
+      "homepage": "https://github.com/cexll/myclaude",
+      "repository": "https://github.com/cexll/myclaude",
+      "license": "MIT",
+      "keywords": [
+        "codex",
+        "code-analysis",
+        "refactoring",
+        "automation",
+        "gpt-5",
+        "ai-coding"
+      ],
+      "category": "essentials",
+      "strict": false,
+      "skills": [
+        "./SKILL.md"
+      ]
+    },
+    {
+      "name": "gemini-cli",
+      "source": "./skills/gemini/",
+      "description": "Execute Gemini CLI for AI-powered code analysis and generation with Google's latest Gemini models",
+      "version": "1.0.0",
+      "author": {
+        "name": "Claude Code Dev Workflows",
+        "url": "https://github.com/cexll/myclaude"
+      },
+      "homepage": "https://github.com/cexll/myclaude",
+      "repository": "https://github.com/cexll/myclaude",
+      "license": "MIT",
+      "keywords": [
+        "gemini",
+        "google-ai",
+        "code-analysis",
+        "code-generation",
+        "ai-reasoning"
+      ],
+      "category": "essentials",
+      "strict": false,
+      "skills": [
+        "./SKILL.md"
+      ]
+    },
+    {
+      "name": "dev-workflow",
+      "source": "./dev-workflow/",
+      "description": "Minimal lightweight development workflow with requirements clarification, parallel codex execution, and mandatory 90% test coverage",
+      "version": "1.0.0",
+      "author": {
+        "name": "Claude Code Dev Workflows",
+        "url": "https://github.com/cexll/myclaude"
+      },
+      "homepage": "https://github.com/cexll/myclaude",
+      "repository": "https://github.com/cexll/myclaude",
+      "license": "MIT",
+      "keywords": [
+        "dev",
+        "workflow",
+        "codex",
+        "testing",
+        "coverage",
+        "concurrent",
+        "lightweight"
+      ],
+      "category": "workflows",
+      "strict": false,
+      "commands": [
+        "./commands/dev.md"
+      ],
+      "agents": [
+        "./agents/dev-plan-generator.md"
+      ]
     }
   ]
 }

.github/workflows/release.yml

@@ -0,0 +1,104 @@
+name: Release codex-wrapper
+
+on:
+  push:
+    tags:
+      - 'v*'
+
+permissions:
+  contents: write
+
+jobs:
+  test:
+    name: Test
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+
+      - name: Setup Go
+        uses: actions/setup-go@v5
+        with:
+          go-version: '1.21'
+
+      - name: Run tests
+        working-directory: codex-wrapper
+        run: go test -v -coverprofile=cover.out ./...
+
+      - name: Check coverage
+        working-directory: codex-wrapper
+        run: |
+          go tool cover -func=cover.out | grep total
+          COVERAGE=$(go tool cover -func=cover.out | grep total | awk '{print $3}' | sed 's/%//')
+          echo "Coverage: ${COVERAGE}%"
+
+  build:
+    name: Build
+    needs: test
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        include:
+          - goos: linux
+            goarch: amd64
+          - goos: linux
+            goarch: arm64
+          - goos: darwin
+            goarch: amd64
+          - goos: darwin
+            goarch: arm64
+
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+
+      - name: Setup Go
+        uses: actions/setup-go@v5
+        with:
+          go-version: '1.21'
+
+      - name: Build binary
+        working-directory: codex-wrapper
+        env:
+          GOOS: ${{ matrix.goos }}
+          GOARCH: ${{ matrix.goarch }}
+          CGO_ENABLED: 0
+        run: |
+          VERSION=${GITHUB_REF#refs/tags/}
+          OUTPUT_NAME=codex-wrapper-${{ matrix.goos }}-${{ matrix.goarch }}
+          go build -ldflags="-s -w -X main.version=${VERSION}" -o ${OUTPUT_NAME} .
+          chmod +x ${OUTPUT_NAME}
+
+      - name: Upload artifact
+        uses: actions/upload-artifact@v4
+        with:
+          name: codex-wrapper-${{ matrix.goos }}-${{ matrix.goarch }}
+          path: codex-wrapper/codex-wrapper-${{ matrix.goos }}-${{ matrix.goarch }}
+
+  release:
+    name: Create Release
+    needs: build
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+
+      - name: Download all artifacts
+        uses: actions/download-artifact@v4
+        with:
+          path: artifacts
+
+      - name: Prepare release files
+        run: |
+          mkdir -p release
+          find artifacts -type f -name "codex-wrapper-*" -exec mv {} release/ \;
+          cp install.sh release/
+          ls -la release/
+
+      - name: Create Release
+        uses: softprops/action-gh-release@v2
+        with:
+          files: release/*
+          generate_release_notes: true
+          draft: false
+          prerelease: false

.gitignore

@@ -1,4 +1,2 @@
-CLAUDE.md
 .claude/
-
-
+.claude-trace

PLUGIN_README.md

@@ -1,95 +0,0 @@
-# Claude Code Plugin System
-
-本项目已支持Claude Code插件系统，可以将命令和代理打包成可安装的插件包。
-
-## 插件配置
-
-插件配置文件位于 `.claude-plugin/marketplace.json`，定义了所有可用的插件包。
-
-## 可用插件
-
-### 1. Requirements-Driven Development
-- **描述**: 需求驱动的开发工作流，包含90%质量门控
-- **命令**: `/requirements-pilot`
-- **代理**: requirements-generate, requirements-code, requirements-testing, requirements-review
-
-### 2. BMAD Agile Workflow
-- **描述**: 完整的BMAD敏捷工作流（产品负责人→架构师→SM→开发→QA）
-- **命令**: `/bmad-pilot`
-- **代理**: bmad-po, bmad-architect, bmad-sm, bmad-dev, bmad-qa, bmad-orchestrator
-
-### 3. Development Essentials
-- **描述**: 核心开发命令套件
-- **命令**: `/code`, `/debug`, `/test`, `/optimize`, `/review`, `/bugfix`, `/refactor`, `/docs`, `/ask`, `/think`
-- **代理**: code, bugfix, bugfix-verify, code-optimize, debug, develop
-
-### 4. Advanced AI Agents
-- **描述**: 高级AI代理，集成GPT-5进行深度分析
-- **代理**: gpt5
-
-## 使用插件命令
-
-### 列出所有可用插件
-```bash
-/plugin list
-```
-
-### 查看插件详情
-```bash
-/plugin info <plugin-name>
-```
-例如：`/plugin info requirements-driven-development`
-
-### 安装插件
-```bash
-/plugin install <plugin-name>
-```
-例如：`/plugin install bmad-agile-workflow`
-
-### 移除插件
-```bash
-/plugin remove <plugin-name>
-```
-
-## 创建自定义插件
-
-要创建自己的插件：
-
-1. 在 `.claude-plugin/marketplace.json` 中添加新的插件定义
-2. 指定插件包含的命令和代理文件路径
-3. 设置适当的元数据（版本、作者、关键词等）
-
-示例插件结构：
-```json
-{
-  "name": "my-custom-plugin",
-  "source": "./",
-  "description": "自定义插件描述",
-  "version": "1.0.0",
-  "commands": [
-    "./commands/my-command.md"
-  ],
-  "agents": [
-    "./agents/my-agent.md"
-  ]
-}
-```
-
-## 分享插件
-
-要分享插件给其他项目：
-1. 复制整个 `.claude-plugin` 目录到目标项目
-2. 确保相关的命令和代理文件存在
-3. 在新项目中使用 `/plugin` 命令管理插件
-
-## 注意事项
-
-- 插件系统遵循Claude Code的插件规范
-- 所有命令和代理文件必须是有效的Markdown格式
-- 插件配置支持版本管理和依赖关系
-- 插件可以包含多个命令、代理和输出样式
-
-## 相关文档
-
-- [Claude Code插件文档](https://docs.claude.com/en/docs/claude-code/plugins)
-- [示例插件仓库](https://github.com/wshobson/agents)

README.md

@@ -2,7 +2,7 @@
 
 [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
 [![Claude Code](https://img.shields.io/badge/Claude-Code-blue)](https://claude.ai/code)
-[![Version](https://img.shields.io/badge/Version-3.2-green)](https://github.com/cexll/myclaude)
+[![Version](https://img.shields.io/badge/Version-4.4-green)](https://github.com/cexll/myclaude)
 [![Plugin Ready](https://img.shields.io/badge/Plugin-Ready-purple)](https://docs.claude.com/en/docs/claude-code/plugins)
 
 > Enterprise-grade agile development automation with AI-powered multi-agent orchestration
@@ -15,7 +15,7 @@
 
 **Plugin System (Recommended)**
 ```bash
-/plugin github.com/cexll/myclaude
+/plugin marketplace add cexll/myclaude
 ```
 
 **Traditional Installation**
@@ -44,6 +44,8 @@ make install
 |--------|-------------|--------------|
 | **[bmad-agile-workflow](docs/BMAD-WORKFLOW.md)** | Complete BMAD methodology with 6 specialized agents | `/bmad-pilot` |
 | **[requirements-driven-workflow](docs/REQUIREMENTS-WORKFLOW.md)** | Streamlined requirements-to-code workflow | `/requirements-pilot` |
+| **[dev-workflow](dev-workflow/README.md)** | Extreme lightweight end-to-end development workflow | `/dev` |
+| **[codex-wrapper](codex-wrapper/)** | Go binary wrapper for Codex CLI integration | `codex-wrapper` |
 | **[development-essentials](docs/DEVELOPMENT-COMMANDS.md)** | Core development slash commands | `/code` `/debug` `/test` `/optimize` |
 | **[advanced-ai-agents](docs/ADVANCED-AGENTS.md)** | GPT-5 deep reasoning integration | Agent: `gpt5` |
 | **[requirements-clarity](docs/REQUIREMENTS-CLARITY.md)** | Automated requirements clarification with 100-point scoring | Auto-activated skill |
@@ -88,6 +90,11 @@ make install
 
 ## 🛠️ Installation Methods
 
+**Codex Wrapper** (Go binary for Codex CLI)
+```bash
+curl -fsSL https://raw.githubusercontent.com/cexll/myclaude/refs/heads/master/install.sh | bash
+```
+
 **Method 1: Plugin Install** (One command)
 ```bash
 /plugin install bmad-agile-workflow
@@ -101,8 +108,8 @@ make deploy-all          # Everything
 ```
 
 **Method 3: Manual Setup**
-- Copy `/commands/*.md` to `~/.config/claude/commands/`
-- Copy `/agents/*.md` to `~/.config/claude/agents/`
+- Copy `./commands/*.md` to `~/.config/claude/commands/`
+- Copy `./agents/*.md` to `~/.config/claude/agents/`
 
 Run `make help` for all options.
 

README_CN.md

@@ -2,7 +2,7 @@
 
 [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
 [![Claude Code](https://img.shields.io/badge/Claude-Code-blue)](https://claude.ai/code)
-[![Version](https://img.shields.io/badge/Version-3.2-green)](https://github.com/cexll/myclaude)
+[![Version](https://img.shields.io/badge/Version-4.4-green)](https://github.com/cexll/myclaude)
 [![Plugin Ready](https://img.shields.io/badge/Plugin-Ready-purple)](https://docs.claude.com/en/docs/claude-code/plugins)
 
 > 企业级敏捷开发自动化与 AI 驱动的多智能体编排
@@ -15,7 +15,7 @@
 
 **插件系统（推荐）**
 ```bash
-/plugin github.com/cexll/myclaude
+/plugin marketplace add cexll/myclaude
 ```
 
 **传统安装**
@@ -44,6 +44,7 @@ make install
 |------|------|---------|
 | **[bmad-agile-workflow](docs/BMAD-WORKFLOW.md)** | 完整 BMAD 方法论，包含6个专业智能体 | `/bmad-pilot` |
 | **[requirements-driven-workflow](docs/REQUIREMENTS-WORKFLOW.md)** | 精简的需求到代码工作流 | `/requirements-pilot` |
+| **[dev-workflow](dev-workflow/README.md)** | 极简端到端开发工作流 | `/dev` |
 | **[development-essentials](docs/DEVELOPMENT-COMMANDS.md)** | 核心开发斜杠命令 | `/code` `/debug` `/test` `/optimize` |
 | **[advanced-ai-agents](docs/ADVANCED-AGENTS.md)** | GPT-5 深度推理集成 | 智能体: `gpt5` |
 | **[requirements-clarity](docs/REQUIREMENTS-CLARITY.md)** | 自动需求澄清，100分制质量评分 | 自动激活技能 |
@@ -101,8 +102,8 @@ make deploy-all          # 全部安装
 ```
 
 **方式3: 手动安装**
-- 复制 `/commands/*.md` 到 `~/.config/claude/commands/`
-- 复制 `/agents/*.md` 到 `~/.config/claude/agents/`
+- 复制 `./commands/*.md` 到 `~/.config/claude/commands/`
+- 复制 `./agents/*.md` 到 `~/.config/claude/agents/`
 
 运行 `make help` 查看所有选项。
 

codex-wrapper/go.mod

@@ -0,0 +1,3 @@
+module codex-wrapper
+
+go 1.25.3

codex-wrapper/main.go

@@ -0,0 +1,905 @@
+package main
+
+import (
+	"bufio"
+	"bytes"
+	"context"
+	"encoding/json"
+	"fmt"
+	"io"
+	"os"
+	"os/exec"
+	"os/signal"
+	"sort"
+	"strconv"
+	"strings"
+	"sync"
+	"syscall"
+	"time"
+)
+
+const (
+	version           = "1.0.0"
+	defaultWorkdir    = "."
+	defaultTimeout    = 7200 // seconds
+	forceKillDelay    = 5    // seconds
+	stdinSpecialChars = "\n\\\"'`$"
+)
+
+// Test hooks for dependency injection
+var (
+	stdinReader      io.Reader = os.Stdin
+	isTerminalFn               = defaultIsTerminal
+	codexCommand               = "codex"
+	buildCodexArgsFn           = buildCodexArgs
+	commandContext             = exec.CommandContext
+	jsonMarshal                = json.Marshal
+)
+
+// Config holds CLI configuration
+type Config struct {
+	Mode          string // "new" or "resume"
+	Task          string
+	SessionID     string
+	WorkDir       string
+	ExplicitStdin bool
+	Timeout       int
+}
+
+// ParallelConfig defines the JSON schema for parallel execution
+type ParallelConfig struct {
+	Tasks []TaskSpec `json:"tasks"`
+}
+
+// TaskSpec describes an individual task entry in the parallel config
+type TaskSpec struct {
+	ID           string   `json:"id"`
+	Task         string   `json:"task"`
+	WorkDir      string   `json:"workdir,omitempty"`
+	Dependencies []string `json:"dependencies,omitempty"`
+	SessionID    string   `json:"session_id,omitempty"`
+	Mode         string   `json:"-"`
+	UseStdin     bool     `json:"-"`
+}
+
+// TaskResult captures the execution outcome of a task
+type TaskResult struct {
+	TaskID    string `json:"task_id"`
+	ExitCode  int    `json:"exit_code"`
+	Message   string `json:"message"`
+	SessionID string `json:"session_id"`
+	Error     string `json:"error"`
+}
+
+func parseParallelConfig(data []byte) (*ParallelConfig, error) {
+	trimmed := bytes.TrimSpace(data)
+	if len(trimmed) == 0 {
+		return nil, fmt.Errorf("parallel config is empty")
+	}
+
+	tasks := strings.Split(string(trimmed), "---TASK---")
+	var cfg ParallelConfig
+	seen := make(map[string]struct{})
+
+	for _, taskBlock := range tasks {
+		taskBlock = strings.TrimSpace(taskBlock)
+		if taskBlock == "" {
+			continue
+		}
+
+		parts := strings.SplitN(taskBlock, "---CONTENT---", 2)
+		if len(parts) != 2 {
+			return nil, fmt.Errorf("task block missing ---CONTENT--- separator")
+		}
+
+		meta := strings.TrimSpace(parts[0])
+		content := strings.TrimSpace(parts[1])
+
+		task := TaskSpec{WorkDir: defaultWorkdir}
+		for _, line := range strings.Split(meta, "\n") {
+			line = strings.TrimSpace(line)
+			if line == "" {
+				continue
+			}
+			kv := strings.SplitN(line, ":", 2)
+			if len(kv) != 2 {
+				continue
+			}
+			key := strings.TrimSpace(kv[0])
+			value := strings.TrimSpace(kv[1])
+
+			switch key {
+			case "id":
+				task.ID = value
+			case "workdir":
+				task.WorkDir = value
+			case "session_id":
+				task.SessionID = value
+				task.Mode = "resume"
+			case "dependencies":
+				for _, dep := range strings.Split(value, ",") {
+					dep = strings.TrimSpace(dep)
+					if dep != "" {
+						task.Dependencies = append(task.Dependencies, dep)
+					}
+				}
+			}
+		}
+
+		if task.ID == "" {
+			return nil, fmt.Errorf("task missing id field")
+		}
+		if content == "" {
+			return nil, fmt.Errorf("task %q missing content", task.ID)
+		}
+		if _, exists := seen[task.ID]; exists {
+			return nil, fmt.Errorf("duplicate task id: %s", task.ID)
+		}
+
+		task.Task = content
+		cfg.Tasks = append(cfg.Tasks, task)
+		seen[task.ID] = struct{}{}
+	}
+
+	if len(cfg.Tasks) == 0 {
+		return nil, fmt.Errorf("no tasks found")
+	}
+
+	return &cfg, nil
+}
+
+func topologicalSort(tasks []TaskSpec) ([][]TaskSpec, error) {
+	idToTask := make(map[string]TaskSpec, len(tasks))
+	indegree := make(map[string]int, len(tasks))
+	adj := make(map[string][]string, len(tasks))
+
+	for _, task := range tasks {
+		idToTask[task.ID] = task
+		indegree[task.ID] = 0
+	}
+
+	for _, task := range tasks {
+		for _, dep := range task.Dependencies {
+			if _, ok := idToTask[dep]; !ok {
+				return nil, fmt.Errorf("dependency %q not found for task %q", dep, task.ID)
+			}
+			indegree[task.ID]++
+			adj[dep] = append(adj[dep], task.ID)
+		}
+	}
+
+	queue := make([]string, 0, len(tasks))
+	for _, task := range tasks {
+		if indegree[task.ID] == 0 {
+			queue = append(queue, task.ID)
+		}
+	}
+
+	layers := make([][]TaskSpec, 0)
+	processed := 0
+
+	for len(queue) > 0 {
+		current := queue
+		queue = nil
+		layer := make([]TaskSpec, len(current))
+		for i, id := range current {
+			layer[i] = idToTask[id]
+			processed++
+		}
+		layers = append(layers, layer)
+
+		next := make([]string, 0)
+		for _, id := range current {
+			for _, neighbor := range adj[id] {
+				indegree[neighbor]--
+				if indegree[neighbor] == 0 {
+					next = append(next, neighbor)
+				}
+			}
+		}
+		queue = append(queue, next...)
+	}
+
+	if processed != len(tasks) {
+		cycleIDs := make([]string, 0)
+		for id, deg := range indegree {
+			if deg > 0 {
+				cycleIDs = append(cycleIDs, id)
+			}
+		}
+		sort.Strings(cycleIDs)
+		return nil, fmt.Errorf("cycle detected involving tasks: %s", strings.Join(cycleIDs, ","))
+	}
+
+	return layers, nil
+}
+
+var runCodexTaskFn = func(task TaskSpec, timeout int) TaskResult {
+	if task.WorkDir == "" {
+		task.WorkDir = defaultWorkdir
+	}
+	if task.Mode == "" {
+		task.Mode = "new"
+	}
+	if task.UseStdin || shouldUseStdin(task.Task, false) {
+		task.UseStdin = true
+	}
+
+	return runCodexTask(task, true, timeout)
+}
+
+func executeConcurrent(layers [][]TaskSpec, timeout int) []TaskResult {
+	totalTasks := 0
+	for _, layer := range layers {
+		totalTasks += len(layer)
+	}
+
+	results := make([]TaskResult, 0, totalTasks)
+	failed := make(map[string]TaskResult, totalTasks)
+	resultsCh := make(chan TaskResult, totalTasks)
+
+	for _, layer := range layers {
+		var wg sync.WaitGroup
+		executed := 0
+
+		for _, task := range layer {
+			if skip, reason := shouldSkipTask(task, failed); skip {
+				res := TaskResult{TaskID: task.ID, ExitCode: 1, Error: reason}
+				results = append(results, res)
+				failed[task.ID] = res
+				continue
+			}
+
+			executed++
+			wg.Add(1)
+			go func(ts TaskSpec) {
+				defer wg.Done()
+				defer func() {
+					if r := recover(); r != nil {
+						resultsCh <- TaskResult{TaskID: ts.ID, ExitCode: 1, Error: fmt.Sprintf("panic: %v", r)}
+					}
+				}()
+				resultsCh <- runCodexTaskFn(ts, timeout)
+			}(task)
+		}
+
+		wg.Wait()
+
+		for i := 0; i < executed; i++ {
+			res := <-resultsCh
+			results = append(results, res)
+			if res.ExitCode != 0 || res.Error != "" {
+				failed[res.TaskID] = res
+			}
+		}
+	}
+
+	return results
+}
+
+func shouldSkipTask(task TaskSpec, failed map[string]TaskResult) (bool, string) {
+	if len(task.Dependencies) == 0 {
+		return false, ""
+	}
+
+	var blocked []string
+	for _, dep := range task.Dependencies {
+		if _, ok := failed[dep]; ok {
+			blocked = append(blocked, dep)
+		}
+	}
+
+	if len(blocked) == 0 {
+		return false, ""
+	}
+
+	return true, fmt.Sprintf("skipped due to failed dependencies: %s", strings.Join(blocked, ","))
+}
+
+func generateFinalOutput(results []TaskResult) string {
+	var sb strings.Builder
+
+	success := 0
+	failed := 0
+	for _, res := range results {
+		if res.ExitCode == 0 && res.Error == "" {
+			success++
+		} else {
+			failed++
+		}
+	}
+
+	sb.WriteString(fmt.Sprintf("=== Parallel Execution Summary ===\n"))
+	sb.WriteString(fmt.Sprintf("Total: %d | Success: %d | Failed: %d\n\n", len(results), success, failed))
+
+	for _, res := range results {
+		sb.WriteString(fmt.Sprintf("--- Task: %s ---\n", res.TaskID))
+		if res.Error != "" {
+			sb.WriteString(fmt.Sprintf("Status: FAILED (exit code %d)\nError: %s\n", res.ExitCode, res.Error))
+		} else if res.ExitCode != 0 {
+			sb.WriteString(fmt.Sprintf("Status: FAILED (exit code %d)\n", res.ExitCode))
+		} else {
+			sb.WriteString("Status: SUCCESS\n")
+		}
+		if res.SessionID != "" {
+			sb.WriteString(fmt.Sprintf("Session: %s\n", res.SessionID))
+		}
+		if res.Message != "" {
+			sb.WriteString(fmt.Sprintf("\n%s\n", res.Message))
+		}
+		sb.WriteString("\n")
+	}
+
+	return sb.String()
+}
+
+// JSONEvent represents a Codex JSON output event
+type JSONEvent struct {
+	Type     string     `json:"type"`
+	ThreadID string     `json:"thread_id,omitempty"`
+	Item     *EventItem `json:"item,omitempty"`
+}
+
+// EventItem represents the item field in a JSON event
+type EventItem struct {
+	Type string      `json:"type"`
+	Text interface{} `json:"text"`
+}
+
+func main() {
+	exitCode := run()
+	os.Exit(exitCode)
+}
+
+// run is the main logic, returns exit code for testability
+func run() int {
+	// Handle --version and --help first
+	if len(os.Args) > 1 {
+		switch os.Args[1] {
+		case "--version", "-v":
+			fmt.Printf("codex-wrapper version %s\n", version)
+			return 0
+		case "--help", "-h":
+			printHelp()
+			return 0
+		case "--parallel":
+			// Parallel mode: read task config from stdin
+			data, err := io.ReadAll(stdinReader)
+			if err != nil {
+				fmt.Fprintf(os.Stderr, "ERROR: failed to read stdin: %v\n", err)
+				return 1
+			}
+
+			cfg, err := parseParallelConfig(data)
+			if err != nil {
+				fmt.Fprintf(os.Stderr, "ERROR: %v\n", err)
+				return 1
+			}
+
+			timeoutSec := resolveTimeout()
+			layers, err := topologicalSort(cfg.Tasks)
+			if err != nil {
+				fmt.Fprintf(os.Stderr, "ERROR: %v\n", err)
+				return 1
+			}
+
+			results := executeConcurrent(layers, timeoutSec)
+			fmt.Println(generateFinalOutput(results))
+
+			exitCode := 0
+			for _, res := range results {
+				if res.ExitCode != 0 {
+					exitCode = res.ExitCode
+				}
+			}
+
+			return exitCode
+		}
+	}
+
+	logInfo("Script started")
+
+	cfg, err := parseArgs()
+	if err != nil {
+		logError(err.Error())
+		return 1
+	}
+	logInfo(fmt.Sprintf("Parsed args: mode=%s, task_len=%d", cfg.Mode, len(cfg.Task)))
+
+	timeoutSec := resolveTimeout()
+	logInfo(fmt.Sprintf("Timeout: %ds", timeoutSec))
+	cfg.Timeout = timeoutSec
+
+	// Determine task text and stdin mode
+	var taskText string
+	var piped bool
+
+	if cfg.ExplicitStdin {
+		logInfo("Explicit stdin mode: reading task from stdin")
+		data, err := io.ReadAll(stdinReader)
+		if err != nil {
+			logError("Failed to read stdin: " + err.Error())
+			return 1
+		}
+		taskText = string(data)
+		if taskText == "" {
+			logError("Explicit stdin mode requires task input from stdin")
+			return 1
+		}
+		piped = !isTerminal()
+	} else {
+		pipedTask := readPipedTask()
+		piped = pipedTask != ""
+		if piped {
+			taskText = pipedTask
+		} else {
+			taskText = cfg.Task
+		}
+	}
+
+	useStdin := cfg.ExplicitStdin || shouldUseStdin(taskText, piped)
+
+	if useStdin {
+		var reasons []string
+		if piped {
+			reasons = append(reasons, "piped input")
+		}
+		if cfg.ExplicitStdin {
+			reasons = append(reasons, "explicit \"-\"")
+		}
+		if strings.Contains(taskText, "\n") {
+			reasons = append(reasons, "newline")
+		}
+		if strings.Contains(taskText, "\\") {
+			reasons = append(reasons, "backslash")
+		}
+		if strings.Contains(taskText, "\"") {
+			reasons = append(reasons, "double-quote")
+		}
+		if strings.Contains(taskText, "'") {
+			reasons = append(reasons, "single-quote")
+		}
+		if strings.Contains(taskText, "`") {
+			reasons = append(reasons, "backtick")
+		}
+		if strings.Contains(taskText, "$") {
+			reasons = append(reasons, "dollar")
+		}
+		if len(taskText) > 800 {
+			reasons = append(reasons, "length>800")
+		}
+		if len(reasons) > 0 {
+			logWarn(fmt.Sprintf("Using stdin mode for task due to: %s", strings.Join(reasons, ", ")))
+		}
+	}
+
+	logInfo("codex running...")
+
+	taskSpec := TaskSpec{
+		Task:      taskText,
+		WorkDir:   cfg.WorkDir,
+		Mode:      cfg.Mode,
+		SessionID: cfg.SessionID,
+		UseStdin:  useStdin,
+	}
+
+	result := runCodexTask(taskSpec, false, cfg.Timeout)
+
+	if result.ExitCode != 0 {
+		return result.ExitCode
+	}
+
+	// Output agent_message
+	fmt.Println(result.Message)
+
+	// Output session_id if present
+	if result.SessionID != "" {
+		fmt.Printf("\n---\nSESSION_ID: %s\n", result.SessionID)
+	}
+
+	return 0
+}
+
+func parseArgs() (*Config, error) {
+	args := os.Args[1:]
+	if len(args) == 0 {
+		return nil, fmt.Errorf("task required")
+	}
+
+	cfg := &Config{
+		WorkDir: defaultWorkdir,
+	}
+
+	// Check for resume mode
+	if args[0] == "resume" {
+		if len(args) < 3 {
+			return nil, fmt.Errorf("resume mode requires: resume <session_id> <task>")
+		}
+		cfg.Mode = "resume"
+		cfg.SessionID = args[1]
+		cfg.Task = args[2]
+		cfg.ExplicitStdin = (args[2] == "-")
+		if len(args) > 3 {
+			cfg.WorkDir = args[3]
+		}
+	} else {
+		cfg.Mode = "new"
+		cfg.Task = args[0]
+		cfg.ExplicitStdin = (args[0] == "-")
+		if len(args) > 1 {
+			cfg.WorkDir = args[1]
+		}
+	}
+
+	return cfg, nil
+}
+
+func readPipedTask() string {
+	if isTerminal() {
+		logInfo("Stdin is tty, skipping pipe read")
+		return ""
+	}
+	logInfo("Reading from stdin pipe...")
+	data, err := io.ReadAll(stdinReader)
+	if err != nil || len(data) == 0 {
+		logInfo("Stdin pipe returned empty data")
+		return ""
+	}
+	logInfo(fmt.Sprintf("Read %d bytes from stdin pipe", len(data)))
+	return string(data)
+}
+
+func shouldUseStdin(taskText string, piped bool) bool {
+	if piped {
+		return true
+	}
+	if len(taskText) > 800 {
+		return true
+	}
+	return strings.IndexAny(taskText, stdinSpecialChars) >= 0
+}
+
+func buildCodexArgs(cfg *Config, targetArg string) []string {
+	if cfg.Mode == "resume" {
+		return []string{
+			"e",
+			"--skip-git-repo-check",
+			"--json",
+			"resume",
+			cfg.SessionID,
+			targetArg,
+		}
+	}
+	return []string{
+		"e",
+		"--skip-git-repo-check",
+		"-C", cfg.WorkDir,
+		"--json",
+		targetArg,
+	}
+}
+
+func runCodexTask(taskSpec TaskSpec, silent bool, timeoutSec int) TaskResult {
+	result := TaskResult{
+		TaskID: taskSpec.ID,
+	}
+
+	cfg := &Config{
+		Mode:      taskSpec.Mode,
+		Task:      taskSpec.Task,
+		SessionID: taskSpec.SessionID,
+		WorkDir:   taskSpec.WorkDir,
+	}
+	if cfg.Mode == "" {
+		cfg.Mode = "new"
+	}
+	if cfg.WorkDir == "" {
+		cfg.WorkDir = defaultWorkdir
+	}
+
+	useStdin := taskSpec.UseStdin
+	targetArg := taskSpec.Task
+	if useStdin {
+		targetArg = "-"
+	}
+
+	codexArgs := buildCodexArgsFn(cfg, targetArg)
+
+	logInfoFn := logInfo
+	logWarnFn := logWarn
+	logErrorFn := logError
+	stderrWriter := io.Writer(os.Stderr)
+	if silent {
+		logInfoFn = func(string) {}
+		logWarnFn = func(string) {}
+		logErrorFn = func(string) {}
+		stderrWriter = io.Discard
+	}
+
+	ctx, cancel := context.WithTimeout(context.Background(), time.Duration(timeoutSec)*time.Second)
+	defer cancel()
+
+	cmd := commandContext(ctx, codexCommand, codexArgs...)
+	cmd.Stderr = stderrWriter
+
+	// Setup stdin if needed
+	var stdinPipe io.WriteCloser
+	var err error
+	if useStdin {
+		stdinPipe, err = cmd.StdinPipe()
+		if err != nil {
+			logErrorFn("Failed to create stdin pipe: " + err.Error())
+			result.ExitCode = 1
+			result.Error = "failed to create stdin pipe: " + err.Error()
+			return result
+		}
+	}
+
+	// Setup stdout
+	stdout, err := cmd.StdoutPipe()
+	if err != nil {
+		logErrorFn("Failed to create stdout pipe: " + err.Error())
+		result.ExitCode = 1
+		result.Error = "failed to create stdout pipe: " + err.Error()
+		return result
+	}
+
+	logInfoFn(fmt.Sprintf("Starting codex with args: codex %s...", strings.Join(codexArgs[:min(5, len(codexArgs))], " ")))
+
+	// Start process
+	if err := cmd.Start(); err != nil {
+		if strings.Contains(err.Error(), "executable file not found") {
+			logErrorFn("codex command not found in PATH")
+			result.ExitCode = 127
+			result.Error = "codex command not found in PATH"
+			return result
+		}
+		logErrorFn("Failed to start codex: " + err.Error())
+		result.ExitCode = 1
+		result.Error = "failed to start codex: " + err.Error()
+		return result
+	}
+	logInfoFn(fmt.Sprintf("Process started with PID: %d", cmd.Process.Pid))
+
+	// Write to stdin if needed
+	if useStdin && stdinPipe != nil {
+		logInfoFn(fmt.Sprintf("Writing %d chars to stdin...", len(taskSpec.Task)))
+		go func() {
+			defer stdinPipe.Close()
+			io.WriteString(stdinPipe, taskSpec.Task)
+		}()
+		logInfoFn("Stdin closed")
+	}
+
+	forwardSignals(ctx, cmd, logErrorFn)
+
+	logInfoFn("Reading stdout...")
+
+	// Parse JSON stream
+	message, threadID := parseJSONStreamWithWarn(stdout, logWarnFn)
+
+	// Wait for process to complete
+	err = cmd.Wait()
+
+	// Check for timeout
+	if ctx.Err() == context.DeadlineExceeded {
+		logErrorFn("Codex execution timeout")
+		if cmd.Process != nil {
+			cmd.Process.Kill()
+		}
+		result.ExitCode = 124
+		result.Error = "codex execution timeout"
+		return result
+	}
+
+	// Check exit code
+	if err != nil {
+		if exitErr, ok := err.(*exec.ExitError); ok {
+			code := exitErr.ExitCode()
+			logErrorFn(fmt.Sprintf("Codex exited with status %d", code))
+			result.ExitCode = code
+			result.Error = fmt.Sprintf("codex exited with status %d", code)
+			return result
+		}
+		logErrorFn("Codex error: " + err.Error())
+		result.ExitCode = 1
+		result.Error = "codex error: " + err.Error()
+		return result
+	}
+
+	if message == "" {
+		logErrorFn("Codex completed without agent_message output")
+		result.ExitCode = 1
+		result.Error = "codex completed without agent_message output"
+		return result
+	}
+
+	result.ExitCode = 0
+	result.Message = message
+	result.SessionID = threadID
+
+	return result
+}
+
+func forwardSignals(ctx context.Context, cmd *exec.Cmd, logErrorFn func(string)) {
+	sigCh := make(chan os.Signal, 1)
+	signal.Notify(sigCh, syscall.SIGINT, syscall.SIGTERM)
+
+	go func() {
+		defer signal.Stop(sigCh)
+		select {
+		case sig := <-sigCh:
+			logErrorFn(fmt.Sprintf("Received signal: %v", sig))
+			if cmd.Process != nil {
+				cmd.Process.Signal(syscall.SIGTERM)
+				time.AfterFunc(time.Duration(forceKillDelay)*time.Second, func() {
+					if cmd.Process != nil {
+						cmd.Process.Kill()
+					}
+				})
+			}
+		case <-ctx.Done():
+		}
+	}()
+}
+
+func parseJSONStream(r io.Reader) (message, threadID string) {
+	return parseJSONStreamWithWarn(r, logWarn)
+}
+
+func parseJSONStreamWithWarn(r io.Reader, warnFn func(string)) (message, threadID string) {
+	scanner := bufio.NewScanner(r)
+	scanner.Buffer(make([]byte, 64*1024), 10*1024*1024)
+
+	if warnFn == nil {
+		warnFn = func(string) {}
+	}
+
+	for scanner.Scan() {
+		line := strings.TrimSpace(scanner.Text())
+		if line == "" {
+			continue
+		}
+
+		var event JSONEvent
+		if err := json.Unmarshal([]byte(line), &event); err != nil {
+			warnFn(fmt.Sprintf("Failed to parse line: %s", truncate(line, 100)))
+			continue
+		}
+
+		// Capture thread_id
+		if event.Type == "thread.started" {
+			threadID = event.ThreadID
+		}
+
+		// Capture agent_message
+		if event.Type == "item.completed" && event.Item != nil && event.Item.Type == "agent_message" {
+			if text := normalizeText(event.Item.Text); text != "" {
+				message = text
+			}
+		}
+	}
+
+	if err := scanner.Err(); err != nil && err != io.EOF {
+		warnFn("Read stdout error: " + err.Error())
+	}
+
+	return message, threadID
+}
+
+func normalizeText(text interface{}) string {
+	switch v := text.(type) {
+	case string:
+		return v
+	case []interface{}:
+		var sb strings.Builder
+		for _, item := range v {
+			if s, ok := item.(string); ok {
+				sb.WriteString(s)
+			}
+		}
+		return sb.String()
+	default:
+		return ""
+	}
+}
+
+func resolveTimeout() int {
+	raw := os.Getenv("CODEX_TIMEOUT")
+	if raw == "" {
+		return defaultTimeout
+	}
+
+	parsed, err := strconv.Atoi(raw)
+	if err != nil || parsed <= 0 {
+		logWarn(fmt.Sprintf("Invalid CODEX_TIMEOUT '%s', falling back to %ds", raw, defaultTimeout))
+		return defaultTimeout
+	}
+
+	// Environment variable is in milliseconds if > 10000, convert to seconds
+	if parsed > 10000 {
+		return parsed / 1000
+	}
+	return parsed
+}
+
+func defaultIsTerminal() bool {
+	fi, err := os.Stdin.Stat()
+	if err != nil {
+		return true
+	}
+	return (fi.Mode() & os.ModeCharDevice) != 0
+}
+
+func isTerminal() bool {
+	return isTerminalFn()
+}
+
+func getEnv(key, defaultValue string) string {
+	if val := os.Getenv(key); val != "" {
+		return val
+	}
+	return defaultValue
+}
+
+func truncate(s string, maxLen int) string {
+	if len(s) <= maxLen {
+		return s
+	}
+	return s[:maxLen] + "..."
+}
+
+func min(a, b int) int {
+	if a < b {
+		return a
+	}
+	return b
+}
+
+func hello() string {
+	return "hello world"
+}
+
+func greet(name string) string {
+	return "hello " + name
+}
+
+func farewell(name string) string {
+	return "goodbye " + name
+}
+
+func logInfo(msg string) {
+	fmt.Fprintf(os.Stderr, "INFO: %s\n", msg)
+}
+
+func logWarn(msg string) {
+	fmt.Fprintf(os.Stderr, "WARN: %s\n", msg)
+}
+
+func logError(msg string) {
+	fmt.Fprintf(os.Stderr, "ERROR: %s\n", msg)
+}
+
+func printHelp() {
+	help := `codex-wrapper - Go wrapper for Codex CLI
+
+Usage:
+    codex-wrapper "task" [workdir]
+    codex-wrapper - [workdir]              Read task from stdin
+    codex-wrapper resume <session_id> "task" [workdir]
+    codex-wrapper resume <session_id> - [workdir]
+    codex-wrapper --version
+    codex-wrapper --help
+
+Environment Variables:
+    CODEX_TIMEOUT  Timeout in milliseconds (default: 7200000)
+
+Exit Codes:
+    0    Success
+    1    General error (missing args, no output)
+    124  Timeout
+    127  codex command not found
+    130  Interrupted (Ctrl+C)
+    *    Passthrough from codex process`
+	fmt.Println(help)
+}

codex-wrapper/main_integration_test.go

@@ -0,0 +1,400 @@
+package main
+
+import (
+	"bytes"
+	"fmt"
+	"io"
+	"os"
+	"strings"
+	"sync"
+	"sync/atomic"
+	"testing"
+	"time"
+)
+
+type integrationSummary struct {
+	Total   int `json:"total"`
+	Success int `json:"success"`
+	Failed  int `json:"failed"`
+}
+
+type integrationOutput struct {
+	Results []TaskResult       `json:"results"`
+	Summary integrationSummary `json:"summary"`
+}
+
+func captureStdout(t *testing.T, fn func()) string {
+	t.Helper()
+	old := os.Stdout
+	r, w, _ := os.Pipe()
+	os.Stdout = w
+
+	fn()
+
+	w.Close()
+	os.Stdout = old
+
+	var buf bytes.Buffer
+	io.Copy(&buf, r)
+	return buf.String()
+}
+
+func parseIntegrationOutput(t *testing.T, out string) integrationOutput {
+	t.Helper()
+	var payload integrationOutput
+
+	lines := strings.Split(out, "\n")
+	var currentTask *TaskResult
+
+	for _, line := range lines {
+		line = strings.TrimSpace(line)
+		if strings.HasPrefix(line, "Total:") {
+			parts := strings.Split(line, "|")
+			for _, p := range parts {
+				p = strings.TrimSpace(p)
+				if strings.HasPrefix(p, "Total:") {
+					fmt.Sscanf(p, "Total: %d", &payload.Summary.Total)
+				} else if strings.HasPrefix(p, "Success:") {
+					fmt.Sscanf(p, "Success: %d", &payload.Summary.Success)
+				} else if strings.HasPrefix(p, "Failed:") {
+					fmt.Sscanf(p, "Failed: %d", &payload.Summary.Failed)
+				}
+			}
+		} else if strings.HasPrefix(line, "--- Task:") {
+			if currentTask != nil {
+				payload.Results = append(payload.Results, *currentTask)
+			}
+			currentTask = &TaskResult{}
+			currentTask.TaskID = strings.TrimSuffix(strings.TrimPrefix(line, "--- Task: "), " ---")
+		} else if currentTask != nil {
+			if strings.HasPrefix(line, "Status: SUCCESS") {
+				currentTask.ExitCode = 0
+			} else if strings.HasPrefix(line, "Status: FAILED") {
+				if strings.Contains(line, "exit code") {
+					fmt.Sscanf(line, "Status: FAILED (exit code %d)", &currentTask.ExitCode)
+				} else {
+					currentTask.ExitCode = 1
+				}
+			} else if strings.HasPrefix(line, "Error:") {
+				currentTask.Error = strings.TrimPrefix(line, "Error: ")
+			} else if strings.HasPrefix(line, "Session:") {
+				currentTask.SessionID = strings.TrimPrefix(line, "Session: ")
+			} else if line != "" && !strings.HasPrefix(line, "===") && !strings.HasPrefix(line, "---") {
+				if currentTask.Message != "" {
+					currentTask.Message += "\n"
+				}
+				currentTask.Message += line
+			}
+		}
+	}
+
+	if currentTask != nil {
+		payload.Results = append(payload.Results, *currentTask)
+	}
+
+	return payload
+}
+
+func findResultByID(t *testing.T, payload integrationOutput, id string) TaskResult {
+	t.Helper()
+	for _, res := range payload.Results {
+		if res.TaskID == id {
+			return res
+		}
+	}
+	t.Fatalf("result for task %s not found", id)
+	return TaskResult{}
+}
+
+func TestParallelEndToEnd_OrderAndConcurrency(t *testing.T) {
+	defer resetTestHooks()
+	origRun := runCodexTaskFn
+	t.Cleanup(func() {
+		runCodexTaskFn = origRun
+		resetTestHooks()
+	})
+
+	input := `---TASK---
+id: A
+---CONTENT---
+task-a
+---TASK---
+id: B
+dependencies: A
+---CONTENT---
+task-b
+---TASK---
+id: C
+dependencies: B
+---CONTENT---
+task-c
+---TASK---
+id: D
+---CONTENT---
+task-d
+---TASK---
+id: E
+---CONTENT---
+task-e`
+	stdinReader = bytes.NewReader([]byte(input))
+	os.Args = []string{"codex-wrapper", "--parallel"}
+
+	var mu sync.Mutex
+	starts := make(map[string]time.Time)
+	ends := make(map[string]time.Time)
+	var running int64
+	var maxParallel int64
+
+	runCodexTaskFn = func(task TaskSpec, timeout int) TaskResult {
+		start := time.Now()
+		mu.Lock()
+		starts[task.ID] = start
+		mu.Unlock()
+
+		cur := atomic.AddInt64(&running, 1)
+		for {
+			prev := atomic.LoadInt64(&maxParallel)
+			if cur <= prev {
+				break
+			}
+			if atomic.CompareAndSwapInt64(&maxParallel, prev, cur) {
+				break
+			}
+		}
+
+		time.Sleep(40 * time.Millisecond)
+
+		mu.Lock()
+		ends[task.ID] = time.Now()
+		mu.Unlock()
+
+		atomic.AddInt64(&running, -1)
+		return TaskResult{TaskID: task.ID, ExitCode: 0, Message: task.Task}
+	}
+
+	var exitCode int
+	output := captureStdout(t, func() {
+		exitCode = run()
+	})
+
+	if exitCode != 0 {
+		t.Fatalf("run() exit = %d, want 0", exitCode)
+	}
+
+	payload := parseIntegrationOutput(t, output)
+	if payload.Summary.Failed != 0 || payload.Summary.Total != 5 || payload.Summary.Success != 5 {
+		t.Fatalf("unexpected summary: %+v", payload.Summary)
+	}
+
+	aEnd := ends["A"]
+	bStart := starts["B"]
+	cStart := starts["C"]
+	bEnd := ends["B"]
+	if aEnd.IsZero() || bStart.IsZero() || bEnd.IsZero() || cStart.IsZero() {
+		t.Fatalf("missing timestamps, starts=%v ends=%v", starts, ends)
+	}
+	if !aEnd.Before(bStart) && !aEnd.Equal(bStart) {
+		t.Fatalf("B should start after A ends: A_end=%v B_start=%v", aEnd, bStart)
+	}
+	if !bEnd.Before(cStart) && !bEnd.Equal(cStart) {
+		t.Fatalf("C should start after B ends: B_end=%v C_start=%v", bEnd, cStart)
+	}
+
+	dStart := starts["D"]
+	eStart := starts["E"]
+	if dStart.IsZero() || eStart.IsZero() {
+		t.Fatalf("missing D/E start times: %v", starts)
+	}
+	delta := dStart.Sub(eStart)
+	if delta < 0 {
+		delta = -delta
+	}
+	if delta > 25*time.Millisecond {
+		t.Fatalf("D and E should run in parallel, delta=%v", delta)
+	}
+	if maxParallel < 2 {
+		t.Fatalf("expected at least 2 concurrent tasks, got %d", maxParallel)
+	}
+}
+
+func TestParallelCycleDetectionStopsExecution(t *testing.T) {
+	defer resetTestHooks()
+	origRun := runCodexTaskFn
+	runCodexTaskFn = func(task TaskSpec, timeout int) TaskResult {
+		t.Fatalf("task %s should not execute on cycle", task.ID)
+		return TaskResult{}
+	}
+	t.Cleanup(func() {
+		runCodexTaskFn = origRun
+		resetTestHooks()
+	})
+
+	input := `---TASK---
+id: A
+dependencies: B
+---CONTENT---
+a
+---TASK---
+id: B
+dependencies: A
+---CONTENT---
+b`
+	stdinReader = bytes.NewReader([]byte(input))
+	os.Args = []string{"codex-wrapper", "--parallel"}
+
+	exitCode := 0
+	output := captureStdout(t, func() {
+		exitCode = run()
+	})
+
+	if exitCode == 0 {
+		t.Fatalf("cycle should cause non-zero exit, got %d", exitCode)
+	}
+	if strings.TrimSpace(output) != "" {
+		t.Fatalf("expected no JSON output on cycle, got %q", output)
+	}
+}
+
+func TestParallelPartialFailureBlocksDependents(t *testing.T) {
+	defer resetTestHooks()
+	origRun := runCodexTaskFn
+	t.Cleanup(func() {
+		runCodexTaskFn = origRun
+		resetTestHooks()
+	})
+
+	runCodexTaskFn = func(task TaskSpec, timeout int) TaskResult {
+		if task.ID == "A" {
+			return TaskResult{TaskID: "A", ExitCode: 2, Error: "boom"}
+		}
+		return TaskResult{TaskID: task.ID, ExitCode: 0, Message: task.Task}
+	}
+
+	input := `---TASK---
+id: A
+---CONTENT---
+fail
+---TASK---
+id: B
+dependencies: A
+---CONTENT---
+blocked
+---TASK---
+id: D
+---CONTENT---
+ok-d
+---TASK---
+id: E
+---CONTENT---
+ok-e`
+	stdinReader = bytes.NewReader([]byte(input))
+	os.Args = []string{"codex-wrapper", "--parallel"}
+
+	var exitCode int
+	output := captureStdout(t, func() {
+		exitCode = run()
+	})
+
+	payload := parseIntegrationOutput(t, output)
+	if exitCode == 0 {
+		t.Fatalf("expected non-zero exit when a task fails, got %d", exitCode)
+	}
+
+	resA := findResultByID(t, payload, "A")
+	resB := findResultByID(t, payload, "B")
+	resD := findResultByID(t, payload, "D")
+	resE := findResultByID(t, payload, "E")
+
+	if resA.ExitCode == 0 {
+		t.Fatalf("task A should fail, got %+v", resA)
+	}
+	if resB.ExitCode == 0 || !strings.Contains(resB.Error, "dependencies") {
+		t.Fatalf("task B should be skipped due to dependency failure, got %+v", resB)
+	}
+	if resD.ExitCode != 0 || resE.ExitCode != 0 {
+		t.Fatalf("independent tasks should run successfully, D=%+v E=%+v", resD, resE)
+	}
+	if payload.Summary.Failed != 2 || payload.Summary.Total != 4 {
+		t.Fatalf("unexpected summary after partial failure: %+v", payload.Summary)
+	}
+}
+
+func TestParallelTimeoutPropagation(t *testing.T) {
+	defer resetTestHooks()
+	origRun := runCodexTaskFn
+	t.Cleanup(func() {
+		runCodexTaskFn = origRun
+		resetTestHooks()
+		os.Unsetenv("CODEX_TIMEOUT")
+	})
+
+	var receivedTimeout int
+	runCodexTaskFn = func(task TaskSpec, timeout int) TaskResult {
+		receivedTimeout = timeout
+		return TaskResult{TaskID: task.ID, ExitCode: 124, Error: "timeout"}
+	}
+
+	os.Setenv("CODEX_TIMEOUT", "1")
+	input := `---TASK---
+id: T
+---CONTENT---
+slow`
+	stdinReader = bytes.NewReader([]byte(input))
+	os.Args = []string{"codex-wrapper", "--parallel"}
+
+	exitCode := 0
+	output := captureStdout(t, func() {
+		exitCode = run()
+	})
+
+	payload := parseIntegrationOutput(t, output)
+	if receivedTimeout != 1 {
+		t.Fatalf("expected timeout 1s to propagate, got %d", receivedTimeout)
+	}
+	if exitCode != 124 {
+		t.Fatalf("expected timeout exit code 124, got %d", exitCode)
+	}
+	if payload.Summary.Failed != 1 || payload.Summary.Total != 1 {
+		t.Fatalf("unexpected summary for timeout case: %+v", payload.Summary)
+	}
+	res := findResultByID(t, payload, "T")
+	if res.Error == "" || res.ExitCode != 124 {
+		t.Fatalf("timeout result not propagated, got %+v", res)
+	}
+}
+
+func TestConcurrentSpeedupBenchmark(t *testing.T) {
+	defer resetTestHooks()
+	origRun := runCodexTaskFn
+	t.Cleanup(func() {
+		runCodexTaskFn = origRun
+		resetTestHooks()
+	})
+
+	runCodexTaskFn = func(task TaskSpec, timeout int) TaskResult {
+		time.Sleep(50 * time.Millisecond)
+		return TaskResult{TaskID: task.ID}
+	}
+
+	tasks := make([]TaskSpec, 10)
+	for i := range tasks {
+		tasks[i] = TaskSpec{ID: fmt.Sprintf("task-%d", i)}
+	}
+	layers := [][]TaskSpec{tasks}
+
+	serialStart := time.Now()
+	for _, task := range tasks {
+		_ = runCodexTaskFn(task, 5)
+	}
+	serialElapsed := time.Since(serialStart)
+
+	concurrentStart := time.Now()
+	_ = executeConcurrent(layers, 5)
+	concurrentElapsed := time.Since(concurrentStart)
+
+	if concurrentElapsed >= serialElapsed/5 {
+		t.Fatalf("expected concurrent time <20%% of serial, serial=%v concurrent=%v", serialElapsed, concurrentElapsed)
+	}
+	ratio := float64(concurrentElapsed) / float64(serialElapsed)
+	t.Logf("speedup ratio (concurrent/serial)=%.3f", ratio)
+}

dev-workflow/README.md

@@ -0,0 +1,163 @@
+# /dev - Minimal Dev Workflow
+
+## Overview
+
+A freshly designed lightweight development workflow with no legacy baggage, focused on delivering high-quality code fast.
+
+## Flow
+
+```
+/dev trigger
+  ↓
+AskUserQuestion (requirements clarification)
+  ↓
+Codex analysis (extract key points and tasks)
+  ↓
+develop-doc-generator (create dev doc)
+  ↓
+Codex concurrent development (2–5 tasks)
+  ↓
+Codex testing & verification (≥90% coverage)
+  ↓
+Done (generate summary)
+```
+
+## The 6 Steps
+
+### 1. Clarify Requirements
+- Use **AskUserQuestion** to ask the user directly
+- No scoring system, no complex logic
+- 2–3 rounds of Q&A until the requirement is clear
+
+### 2. Codex Analysis
+- Call codex to analyze the request
+- Extract: core functions, technical points, task list (2–5 items)
+- Output a structured analysis
+
+### 3. Generate Dev Doc
+- Call the **develop-doc-generator** agent
+- Produce a single `dev-plan.md`
+- Include: task breakdown, file scope, dependencies, test commands
+
+### 4. Concurrent Development
+- Work from the task list in dev-plan.md
+- Independent tasks → run in parallel
+- Conflicting tasks → run serially
+
+### 5. Testing & Verification
+- Each codex task:
+  - Implements the feature
+  - Writes tests
+  - Runs coverage
+  - Reports results (≥90%)
+
+### 6. Complete
+- Summarize task status
+- Record coverage
+
+## Usage
+
+```bash
+/dev "Implement user login with email + password"
+```
+
+**No options**, fixed workflow, works out of the box.
+
+## Output Structure
+
+```
+.claude/specs/{feature_name}/
+└──  dev-plan.md      # Dev document generated by agent
+```
+
+Only one file—minimal and clear.
+
+## Core Components
+
+### Tools
+- **AskUserQuestion**: interactive requirement clarification
+- **codex**: analysis, development, testing
+- **develop-doc-generator**: generate dev doc (subagent, saves context)
+
+## Key Features
+
+### ✅ Fresh Design
+- No legacy project residue
+- No complex scoring logic
+- No extra abstraction layers
+
+### ✅ Minimal Orchestration
+- Orchestrator controls the flow directly
+- Only three tools/components
+- Steps are straightforward
+
+### ✅ Concurrency
+- 2–5 tasks in parallel
+- Auto-detect dependencies and conflicts
+- Codex executes independently
+
+### ✅ Quality Assurance
+- Enforces 90% coverage
+- Codex tests and verifies its own work
+- Automatic retry on failure
+
+## Example
+
+```bash
+# Trigger
+/dev "Add user login feature"
+
+# Step 1: Clarify requirements
+Q: What login methods are supported?
+A: Email + password
+Q: Should login be remembered?
+A: Yes, use JWT token
+
+# Step 2: Codex analysis
+Output:
+- Core: email/password login + JWT auth
+- Task 1: Backend API
+- Task 2: Password hashing
+- Task 3: Frontend form
+
+# Step 3: Generate doc
+dev-plan.md generated ✓
+
+# Step 4-5: Concurrent development
+[task-1] Backend API → tests → 92% ✓
+[task-2] Password hashing → tests → 95% ✓
+[task-3] Frontend form → tests → 91% ✓
+```
+
+## Directory Structure
+
+```
+dev-workflow/
+├── README.md                          # This doc
+├── commands/
+│   └── dev.md                         # Workflow definition
+└── agents/
+    └── develop-doc-generator.md       # Doc generator
+```
+
+Minimal structure, only three files.
+
+## When to Use
+
+✅ **Good for**:
+- Any feature size
+- Fast iterations
+- High test coverage needs
+- Wanting concurrent speed-up
+
+## Design Principles
+
+1. **KISS**: keep it simple
+2. **Disposable**: no persistent config
+3. **Quality first**: enforce 90% coverage
+4. **Concurrency first**: leverage codex
+5. **No legacy baggage**: clean-slate design
+
+---
+
+**Philosophy**: zero tolerance for complexity—ship the smallest usable solution, like Linus would.

dev-workflow/agents/dev-plan-generator.md

@@ -0,0 +1,114 @@
+---
+name: dev-plan-generator
+description: Use this agent when you need to generate a structured development plan document (`dev-plan.md`) that breaks down a feature into concrete implementation tasks with testing requirements and acceptance criteria. This agent should be called after requirements analysis and before actual implementation begins.\n\n<example>\nContext: User is orchestrating a feature development workflow and needs to create a development plan after Codex analysis is complete.\nuser: "Create a development plan for the user authentication feature based on the requirements and analysis"\nassistant: "I'll use the Task tool to launch the dev-plan-generator agent to create the structured development plan document."\n<commentary>\nThe user needs a dev-plan.md document generated from requirements and analysis. Use the dev-plan-generator agent to create the structured task breakdown.\n</commentary>\n</example>\n\n<example>\nContext: Orchestrator has completed requirements gathering and Codex analysis for a new feature and needs to generate the development plan before moving to implementation.\nuser: "We've completed the analysis for the payment integration feature. Generate the development plan."\nassistant: "I'm going to use the Task tool to launch the dev-plan-generator agent to create the dev-plan.md document with task breakdown and testing requirements."\n<commentary>\nThis is the step in the workflow where the development plan document needs to be generated. Use the dev-plan-generator agent to create the structured plan.\n</commentary>\n</example>\n\n<example>\nContext: User is working through a requirements-driven workflow and has just approved the technical specifications.\nuser: "The specs look good. Let's move forward with creating the implementation plan."\nassistant: "I'll use the Task tool to launch the dev-plan-generator agent to generate the dev-plan.md document with the task breakdown."\n<commentary>\nAfter spec approval, the next step is generating the development plan. Use the dev-plan-generator agent to create the structured document.\n</commentary>\n</example>
+tools: Glob, Grep, Read, Edit, Write, TodoWrite
+model: sonnet
+color: green
+---
+
+You are a specialized Development Plan Document Generator. Your sole responsibility is to create structured, actionable development plan documents (`dev-plan.md`) that break down features into concrete implementation tasks.
+
+## Your Role
+
+You receive context from an orchestrator including:
+- Feature requirements description
+- Codex analysis results (feature highlights, task decomposition)
+- Feature name (in kebab-case format)
+
+Your output is a single file: `./.claude/specs/{feature_name}/dev-plan.md`
+
+## Document Structure You Must Follow
+
+```markdown
+# {Feature Name} - Development Plan
+
+## Overview
+[One-sentence description of core functionality]
+
+## Task Breakdown
+
+### Task 1: [Task Name]
+- **ID**: task-1
+- **Description**: [What needs to be done]
+- **File Scope**: [Directories or files involved, e.g., src/auth/**, tests/auth/]
+- **Dependencies**: [None or depends on task-x]
+- **Test Command**: [e.g., pytest tests/auth --cov=src/auth --cov-report=term]
+- **Test Focus**: [Scenarios to cover]
+
+### Task 2: [Task Name]
+...
+
+(2-5 tasks)
+
+## Acceptance Criteria
+- [ ] Feature point 1
+- [ ] Feature point 2
+- [ ] All unit tests pass
+- [ ] Code coverage ≥90%
+
+## Technical Notes
+- [Key technical decisions]
+- [Constraints to be aware of]
+```
+
+## Generation Rules You Must Enforce
+
+1. **Task Count**: Generate 2-5 tasks (no more, no less unless the feature is extremely simple or complex)
+2. **Task Requirements**: Each task MUST include:
+   - Clear ID (task-1, task-2, etc.)
+   - Specific description of what needs to be done
+   - Explicit file scope (directories or files affected)
+   - Dependency declaration ("None" or "depends on task-x")
+   - Complete test command with coverage parameters
+   - Testing focus points (scenarios to cover)
+3. **Task Independence**: Design tasks to be as independent as possible to enable parallel execution
+4. **Test Commands**: Must include coverage parameters (e.g., `--cov=module --cov-report=term` for pytest, `--coverage` for npm)
+5. **Coverage Threshold**: Always require ≥90% code coverage in acceptance criteria
+
+## Your Workflow
+
+1. **Analyze Input**: Review the requirements description and Codex analysis results
+2. **Identify Tasks**: Break down the feature into 2-5 logical, independent tasks
+3. **Determine Dependencies**: Map out which tasks depend on others (minimize dependencies)
+4. **Specify Testing**: For each task, define the exact test command and coverage requirements
+5. **Define Acceptance**: List concrete, measurable acceptance criteria including the 90% coverage requirement
+6. **Document Technical Points**: Note key technical decisions and constraints
+7. **Write File**: Use the Write tool to create `./.claude/specs/{feature_name}/dev-plan.md`
+
+## Quality Checks Before Writing
+
+- [ ] Task count is between 2-5
+- [ ] Every task has all 6 required fields (ID, Description, File Scope, Dependencies, Test Command, Test Focus)
+- [ ] Test commands include coverage parameters
+- [ ] Dependencies are explicitly stated
+- [ ] Acceptance criteria includes 90% coverage requirement
+- [ ] File scope is specific (not vague like "all files")
+- [ ] Testing focus is concrete (not generic like "test everything")
+
+## Critical Constraints
+
+- **Document Only**: You generate documentation. You do NOT execute code, run tests, or modify source files.
+- **Single Output**: You produce exactly one file: `dev-plan.md` in the correct location
+- **Path Accuracy**: The path must be `./.claude/specs/{feature_name}/dev-plan.md` where {feature_name} matches the input
+- **Language Matching**: Output language matches user input (Chinese input → Chinese doc, English input → English doc)
+- **Structured Format**: Follow the exact markdown structure provided
+
+## Example Output Quality
+
+Refer to the user login example in your instructions as the quality benchmark. Your outputs should have:
+- Clear, actionable task descriptions
+- Specific file paths (not generic)
+- Realistic test commands for the actual tech stack
+- Concrete testing scenarios (not abstract)
+- Measurable acceptance criteria
+- Relevant technical decisions
+
+## Error Handling
+
+If the input context is incomplete or unclear:
+1. Request the missing information explicitly
+2. Do NOT proceed with generating a low-quality document
+3. Do NOT make up requirements or technical details
+4. Ask for clarification on: feature scope, tech stack, testing framework, file structure
+
+Remember: Your document will be used by other agents to implement the feature. Precision and completeness are critical. Every field must be filled with specific, actionable information.

dev-workflow/commands/dev.md

@@ -0,0 +1,110 @@
+---
+description: Extreme lightweight end-to-end development workflow with requirements clarification, parallel codex execution, and mandatory 90% test coverage
+---
+
+
+You are the /dev Workflow Orchestrator, an expert development workflow manager specializing in orchestrating minimal, efficient end-to-end development processes with parallel task execution and rigorous test coverage validation.
+
+**Core Responsibilities**
+- Orchestrate a streamlined 6-step development workflow:
+  1. Requirement clarification through targeted questioning
+  2. Technical analysis using Codex
+  3. Development documentation generation
+  4. Parallel development execution
+  5. Coverage validation (≥90% requirement)
+  6. Completion summary
+
+**Workflow Execution**
+- **Step 1: Requirement Clarification**
+  - Use AskUserQuestion to clarify requirements directly
+  - Focus questions on functional boundaries, inputs/outputs, constraints, testing
+  - Iterate 2-3 rounds until clear; rely on judgment; keep questions concise
+
+- **Step 2: Codex Deep Analysis (Plan Mode Style)**
+
+  Use Codex Skill to perform deep analysis. Codex should operate in "plan mode" style:
+
+  **When Deep Analysis is Needed** (any condition triggers):
+  - Multiple valid approaches exist (e.g., Redis vs in-memory vs file-based caching)
+  - Significant architectural decisions required (e.g., WebSockets vs SSE vs polling)
+  - Large-scale changes touching many files or systems
+  - Unclear scope requiring exploration first
+
+  **What Codex Does in Analysis Mode**:
+  1. **Explore Codebase**: Use Glob, Grep, Read to understand structure, patterns, architecture
+  2. **Identify Existing Patterns**: Find how similar features are implemented, reuse conventions
+  3. **Evaluate Options**: When multiple approaches exist, list trade-offs (complexity, performance, security, maintainability)
+  4. **Make Architectural Decisions**: Choose patterns, APIs, data models with justification
+  5. **Design Task Breakdown**: Produce 2-5 parallelizable tasks with file scope and dependencies
+
+  **Analysis Output Structure**:
+  ```
+  ## Context & Constraints
+  [Tech stack, existing patterns, constraints discovered]
+
+  ## Codebase Exploration
+  [Key files, modules, patterns found via Glob/Grep/Read]
+
+  ## Implementation Options (if multiple approaches)
+  | Option | Pros | Cons | Recommendation |
+
+  ## Technical Decisions
+  [API design, data models, architecture choices made]
+
+  ## Task Breakdown
+  [2-5 tasks with: ID, description, file scope, dependencies, test command]
+  ```
+
+  **Skip Deep Analysis When**:
+  - Simple, straightforward implementation with obvious approach
+  - Small changes confined to 1-2 files
+  - Clear requirements with single implementation path
+
+- **Step 3: Generate Development Documentation**
+  - invoke agent dev-plan-generator
+  - Output a brief summary of dev-plan.md:
+    - Number of tasks and their IDs
+    - File scope for each task
+    - Dependencies between tasks
+    - Test commands
+  - Use AskUserQuestion to confirm with user:
+    - Question: "Proceed with this development plan?"
+    - Options: "Confirm and execute" / "Need adjustments"
+  - If user chooses "Need adjustments", return to Step 1 or Step 2 based on feedback
+
+- **Step 4: Parallel Development Execution**
+  - For each task in `dev-plan.md`, invoke Codex with this brief:
+    ```
+    Task: [task-id]
+    Reference: @.claude/specs/{feature_name}/dev-plan.md
+    Scope: [task file scope]
+    Test: [test command]
+    Deliverables: code + unit tests + coverage ≥90% + coverage summary
+    ```
+  - Execute independent tasks concurrently; serialize conflicting ones; track coverage reports
+
+- **Step 5: Coverage Validation**
+  - Validate each task’s coverage:
+    - All ≥90% → pass
+    - Any <90% → request more tests (max 2 rounds)
+
+- **Step 6: Completion Summary**
+  - Provide completed task list, coverage per task, key file changes
+
+**Error Handling**
+- Codex failure: retry once, then log and continue
+- Insufficient coverage: request more tests (max 2 rounds)
+- Dependency conflicts: serialize automatically
+
+**Quality Standards**
+- Code coverage ≥90%
+- 2-5 genuinely parallelizable tasks
+- Documentation must be minimal yet actionable
+- No verbose implementations; only essential code
+
+**Communication Style**
+- Be direct and concise
+- Report progress at each workflow step
+- Highlight blockers immediately
+- Provide actionable next steps when coverage fails
+- Prioritize speed via parallelization while enforcing coverage validation

docs/PLUGIN-SYSTEM.md

@@ -104,7 +104,7 @@ This repository provides 4 ready-to-use Claude Code plugins that can be installe
 
 ```bash
 # Install from GitHub repository
-/plugin github.com/cexll/myclaude
+/plugin marketplace add cexll/myclaude
 ```
 
 This will present all available plugins from the repository.

docs/QUICK-START.md

@@ -8,7 +8,7 @@
 
 ```bash
 # Install everything with one command
-/plugin github.com/cexll/myclaude
+/plugin marketplace add cexll/myclaude
 ```
 
 ### Option 2: Make Install

install.sh

@@ -0,0 +1,46 @@
+#!/bin/bash
+set -e
+
+# Detect platform
+OS=$(uname -s | tr '[:upper:]' '[:lower:]')
+ARCH=$(uname -m)
+
+# Normalize architecture names
+case "$ARCH" in
+    x86_64) ARCH="amd64" ;;
+    aarch64|arm64) ARCH="arm64" ;;
+    *) echo "Unsupported architecture: $ARCH" >&2; exit 1 ;;
+esac
+
+# Build download URL
+REPO="cexll/myclaude"
+VERSION="latest"
+BINARY_NAME="codex-wrapper-${OS}-${ARCH}"
+URL="https://github.com/${REPO}/releases/${VERSION}/download/${BINARY_NAME}"
+
+echo "Downloading codex-wrapper from ${URL}..."
+if ! curl -fsSL "$URL" -o /tmp/codex-wrapper; then
+    echo "ERROR: failed to download binary" >&2
+    exit 1
+fi
+
+mkdir -p "$HOME/bin"
+
+mv /tmp/codex-wrapper "$HOME/bin/codex-wrapper"
+chmod +x "$HOME/bin/codex-wrapper"
+
+if "$HOME/bin/codex-wrapper" --version >/dev/null 2>&1; then
+    echo "codex-wrapper installed successfully to ~/bin/codex-wrapper"
+else
+    echo "ERROR: installation verification failed" >&2
+    exit 1
+fi
+
+if [[ ":$PATH:" != *":$HOME/bin:"* ]]; then
+    echo ""
+    echo "WARNING: ~/bin is not in your PATH"
+    echo "Add this line to your ~/.bashrc or ~/.zshrc:"
+    echo ""
+    echo "    export PATH=\"\$HOME/bin:\$PATH\""
+    echo ""
+fi

memorys/CLAUDE.md

@@ -0,0 +1,61 @@
+You are Linus Torvalds. Obey the following priority stack (highest first) and refuse conflicts by citing the higher rule:
+1. Role + Safety: stay in character, enforce KISS/YAGNI/never break userspace, think in English, respond to the user in Chinese, stay technical.
+2. Workflow Contract: Claude Code performs intake, context gathering, planning, and verification only; every edit or test must be executed via Codex skill (`codex`).
+3. Tooling & Safety Rules:
+   - Capture errors, retry once if transient, document fallbacks.
+4. Context Blocks & Persistence: honor `<context_gathering>`, `<exploration>`, `<persistence>`, `<tool_preambles>`, and `<self_reflection>` exactly as written below.
+5. Quality Rubrics: follow the code-editing rules, implementation checklist, and communication standards; keep outputs concise.
+6. Reporting: summarize in Chinese, include file paths with line numbers, list risks and next steps when relevant.
+
+<context_gathering>
+Fetch project context in parallel: README, package.json/pyproject.toml, directory structure, main configs.
+Method: batch parallel searches, no repeated queries, prefer action over excessive searching.
+Early stop criteria: can name exact files/content to change, or search results 70% converge on one area.
+Budget: 5-8 tool calls, justify overruns.
+</context_gathering>
+
+<exploration>
+Goal: Decompose and map the problem space before planning.
+Trigger conditions:
+- Task involves ≥3 steps or multiple files
+- User explicitly requests deep analysis
+Process:
+- Requirements: Break the ask into explicit requirements, unclear areas, and hidden assumptions.
+- Scope mapping: Identify codebase regions, files, functions, or libraries likely involved. If unknown, perform targeted parallel searches NOW before planning. For complex codebases or deep call chains, delegate scope analysis to Codex skill.
+- Dependencies: Identify relevant frameworks, APIs, config files, data formats, and versioning concerns. When dependencies involve complex framework internals or multi-layer interactions, delegate to Codex skill for analysis.
+- Ambiguity resolution: Choose the most probable interpretation based on repo context, conventions, and dependency docs. Document assumptions explicitly.
+- Output contract: Define exact deliverables (files changed, expected outputs, API responses, CLI behavior, tests passing, etc.).
+In plan mode: Invest extra effort here—this phase determines plan quality and depth.
+</exploration>
+
+<persistence>
+Keep acting until the task is fully solved. Do not hand control back due to uncertainty; choose the most reasonable assumption and proceed.
+If the user asks "should we do X?" and the answer is yes, execute directly without waiting for confirmation.
+Extreme bias for action: when instructions are ambiguous, assume the user wants you to execute rather than ask back.
+</persistence>
+
+<tool_preambles>
+Before any tool call, restate the user goal and outline the current plan. While executing, narrate progress briefly per step. Conclude with a short recap distinct from the upfront plan.
+</tool_preambles>
+
+<self_reflection>
+Construct a private rubric with at least five categories (maintainability, tests with ≥90% coverage, performance, security, style, documentation, backward compatibility). Evaluate the work before finalizing; revisit the implementation if any category misses the bar.
+</self_reflection>
+
+<output_verbosity>
+- Small changes (≤10 lines): 2-5 sentences, no headings, at most 1 short code snippet
+- Medium changes: ≤6 bullet points, at most 2 code snippets (≤8 lines each)
+- Large changes: summarize by file grouping, avoid inline code
+- Do not output build/test logs unless blocking or user requests
+</output_verbosity>
+
+Code Editing Rules:
+- Favor simple, modular solutions; keep indentation ≤3 levels and functions single-purpose.
+- Reuse existing patterns; Tailwind/shadcn defaults for frontend; readable naming over cleverness.
+- Comments only when intent is non-obvious; keep them short.
+- Enforce accessibility, consistent spacing (multiples of 4), ≤2 accent colors.
+- Use semantic HTML and accessible components.
+Communication:
+- Think in English, respond in Chinese, stay terse.
+- Lead with findings before summaries; critique code, not people.
+- Provide next steps only when they naturally follow from the work.

skills/codex/SKILL.md

@@ -14,30 +14,63 @@ Execute Codex CLI commands and parse structured JSON responses. Supports file re
 - Complex code analysis requiring deep understanding
 - Large-scale refactoring across multiple files
 - Automated code generation with safety controls
-- Tasks requiring specialized reasoning models (o3, gpt-5)
+
+## Fallback Policy
+
+Codex is the **primary execution method** for all code edits and tests. Direct execution is only permitted when:
+
+1. Codex is unavailable (service down, network issues)
+2. Codex fails **twice consecutively** on the same task
+
+When falling back to direct execution:
+- Log `CODEX_FALLBACK` with the reason
+- Retry Codex on the next task (don't permanently switch)
+- Document the fallback in the final summary
 
 ## Usage
 
-**推荐方式**（使用 uv run，自动管理 Python 环境）：
+**Mandatory**: Run every automated invocation through the Bash tool in the foreground with **HEREDOC syntax** to avoid shell quoting issues, keeping the `timeout` parameter fixed at `7200000` milliseconds (do not change it or use any other entry point).
+
 ```bash
-uv run ~/.claude/skills/codex/scripts/codex.py "<task>" [model] [working_dir]
+codex-wrapper - [working_dir] <<'EOF'
+<task content here>
+EOF
 ```
 
-**备选方式**（直接执行或使用 Python）：
+**Why HEREDOC?** Tasks often contain code blocks, nested quotes, shell metacharacters (`$`, `` ` ``, `\`), and multiline text. HEREDOC (Here Document) syntax passes these safely without shell interpretation, eliminating quote-escaping nightmares.
+
+**Foreground only (no background/BashOutput)**: Never set `background: true`, never accept Claude's "Running in the background" mode, and avoid `BashOutput` streaming loops. Keep a single foreground Bash call per Codex task; if work might be long, split it into smaller foreground runs instead of offloading to background execution.
+
+**Simple tasks** (backward compatibility):
+For simple single-line tasks without special characters, you can still use direct quoting:
 ```bash
-~/.claude/skills/codex/scripts/codex.py "<task>" [model] [working_dir]
-# 或
-python3 ~/.claude/skills/codex/scripts/codex.py "<task>" [model] [working_dir]
+codex-wrapper "simple task here" [working_dir]
 ```
 
-恢复会话:
+**Resume a session with HEREDOC:**
 ```bash
-uv run ~/.claude/skills/codex/scripts/codex.py resume <session_id> "<task>" [model] [working_dir]
+codex-wrapper resume <session_id> - [working_dir] <<'EOF'
+<task content>
+EOF
 ```
 
+**Cross-platform notes:**
+- **Bash/Zsh**: Use `<<'EOF'` (single quotes prevent variable expansion)
+- **PowerShell 5.1+**: Use `@'` and `'@` (here-string syntax)
+  ```powershell
+  codex-wrapper - @'
+  task content
+  '@
+  ```
+
+## Environment Variables
+
+- **CODEX_TIMEOUT**: Override timeout in milliseconds (default: 7200000 = 2 hours)
+  - Example: `export CODEX_TIMEOUT=3600000` for 1 hour
+
 ## Timeout Control
 
-- **Built-in**: Script enforces 2-hour timeout by default
+- **Built-in**: Binary enforces 2-hour timeout by default
 - **Override**: Set `CODEX_TIMEOUT` environment variable (in milliseconds, e.g., `CODEX_TIMEOUT=3600000` for 1 hour)
 - **Behavior**: On timeout, sends SIGTERM, then SIGKILL after 5s if process doesn't exit
 - **Exit code**: Returns 124 on timeout (consistent with GNU timeout)
@@ -46,9 +79,6 @@ uv run ~/.claude/skills/codex/scripts/codex.py resume <session_id> "<task>" [mod
 ### Parameters
 
 - `task` (required): Task description, supports `@file` references
-- `model` (optional): Model to use (default: gpt-5.1-codex)
-  - `gpt-5.1-codex`: Default, optimized for code
-  - `gpt-5`: Fast general purpose
 - `working_dir` (optional): Working directory (default: current)
 
 ### Return Format
@@ -66,73 +96,167 @@ Error format (stderr):
 ERROR: Error message
 ```
 
+Return only the final agent message and session ID—do not paste raw `BashOutput` logs or background-task chatter into the conversation.
+
 ### Invocation Pattern
 
-When calling via Bash tool, always include the timeout parameter:
+All automated executions must use HEREDOC syntax through the Bash tool in the foreground, with `timeout` fixed at `7200000` (non-negotiable):
+
 ```
 Bash tool parameters:
-- command: uv run ~/.claude/skills/codex/scripts/codex.py "<task>" [model] [working_dir]
+- command: codex-wrapper - [working_dir] <<'EOF'
+  <task content>
+  EOF
 - timeout: 7200000
 - description: <brief description of the task>
 ```
 
-Alternatives:
-```
-# Direct execution (simplest)
-- command: ~/.claude/skills/codex/scripts/codex.py "<task>" [model] [working_dir]
+Run every call in the foreground—never append `&` to background it—so logs and errors stay visible for timely interruption or diagnosis.
 
-# Using python3
-- command: python3 ~/.claude/skills/codex/scripts/codex.py "<task>" [model] [working_dir]
-```
+**Important:** Use HEREDOC (`<<'EOF'`) for all but the simplest tasks. This prevents shell interpretation of quotes, variables, and special characters.
 
 ### Examples
 
 **Basic code analysis:**
 ```bash
-# Recommended: via uv run (auto-manages Python environment)
-uv run ~/.claude/skills/codex/scripts/codex.py "explain @src/main.ts"
+# Recommended: with HEREDOC (handles any special characters)
+codex-wrapper - <<'EOF'
+explain @src/main.ts
+EOF
 # timeout: 7200000
 
-# Alternative: direct execution
-~/.claude/skills/codex/scripts/codex.py "explain @src/main.ts"
+# Alternative: simple direct quoting (if task is simple)
+codex-wrapper "explain @src/main.ts"
 ```
 
-**Refactoring with specific model:**
+**Refactoring with multiline instructions:**
 ```bash
-uv run ~/.claude/skills/codex/scripts/codex.py "refactor @src/utils for performance" "gpt-5"
+codex-wrapper - <<'EOF'
+refactor @src/utils for performance:
+- Extract duplicate code into helpers
+- Use memoization for expensive calculations
+- Add inline comments for non-obvious logic
+EOF
 # timeout: 7200000
 ```
 
 **Multi-file analysis:**
 ```bash
-uv run ~/.claude/skills/codex/scripts/codex.py "analyze @. and find security issues" "gpt-5-codex" "/path/to/project"
+codex-wrapper - "/path/to/project" <<'EOF'
+analyze @. and find security issues:
+1. Check for SQL injection vulnerabilities
+2. Identify XSS risks in templates
+3. Review authentication/authorization logic
+4. Flag hardcoded credentials or secrets
+EOF
 # timeout: 7200000
 ```
 
 **Resume previous session:**
 ```bash
 # First session
-uv run ~/.claude/skills/codex/scripts/codex.py "add comments to @utils.js" "gpt-5-codex"
+codex-wrapper - <<'EOF'
+add comments to @utils.js explaining the caching logic
+EOF
 # Output includes: SESSION_ID: 019a7247-ac9d-71f3-89e2-a823dbd8fd14
 
-# Continue the conversation
-uv run ~/.claude/skills/codex/scripts/codex.py resume 019a7247-ac9d-71f3-89e2-a823dbd8fd14 "now add type hints"
+# Continue the conversation with more context
+codex-wrapper resume 019a7247-ac9d-71f3-89e2-a823dbd8fd14 - <<'EOF'
+now add TypeScript type hints and handle edge cases where cache is null
+EOF
 # timeout: 7200000
 ```
 
-**Using python3 directly (alternative):**
+**Task with code snippets and special characters:**
 ```bash
-python3 ~/.claude/skills/codex/scripts/codex.py "your task here"
+codex-wrapper - <<'EOF'
+Fix the bug in @app.js where the regex /\d+/ doesn't match "123"
+The current code is:
+  const re = /\d+/;
+  if (re.test(input)) { ... }
+Add proper escaping and handle $variables correctly.
+EOF
 ```
 
+### Parallel Execution
+
+For multiple independent or dependent tasks, use `--parallel` mode with delimiter format:
+
+```bash
+codex-wrapper --parallel - <<'EOF'
+---TASK---
+id: analyze_1732876800
+workdir: /home/user/project
+---CONTENT---
+analyze requirements @spec.md
+---TASK---
+id: implement_1732876801
+workdir: /home/user/project
+dependencies: analyze_1732876800
+---CONTENT---
+implement feature based on analyze_1732876800 analysis
+---TASK---
+id: docs_1732876802
+workdir: /home/user/project/docs
+---CONTENT---
+independent task runs in parallel with analyze_1732876800
+EOF
+```
+
+**Delimiter Format**:
+- `---TASK---`: Starts a new task block
+- `id: <task-id>`: Required, unique task identifier
+  - Best practice: use `<feature>_<timestamp>` format (e.g., `auth_1732876800`, `api_test_1732876801`)
+  - Ensures uniqueness across runs and makes tasks traceable
+- `workdir: <path>`: Optional, working directory (default: `.`)
+  - Best practice: use absolute paths (e.g., `/home/user/project/backend`)
+  - Avoids ambiguity and ensures consistent behavior across environments
+- `dependencies: <id1>, <id2>`: Optional, comma-separated task IDs
+- `session_id: <uuid>`: Optional, resume a previous session
+- `---CONTENT---`: Separates metadata from task content
+- Task content: Any text, code, special characters (no escaping needed)
+
+**Resume Failed Tasks**:
+```bash
+# Use session_id from previous output to resume
+codex-wrapper --parallel - <<'EOF'
+---TASK---
+id: T2
+session_id: 019xxx-previous-session-id
+---CONTENT---
+fix the previous error and retry
+EOF
+```
+
+**Output**: Human-readable text format
+```
+=== Parallel Execution Summary ===
+Total: 3 | Success: 2 | Failed: 1
+
+--- Task: T1 ---
+Status: SUCCESS
+Session: 019xxx
+
+Task output message...
+
+--- Task: T2 ---
+Status: FAILED (exit code 1)
+Error: some error message
+```
+
+**Features**:
+- Automatic topological sorting based on dependencies
+- Unlimited concurrency for independent tasks
+- Error isolation (failed tasks don't stop others)
+- Dependency blocking (dependent tasks skip if parent fails)
+
 ## Notes
 
-- **Recommended**: Use `uv run` for automatic Python environment management (requires uv installed)
-- **Alternative**: Direct execution `./codex.py` (uses system Python via shebang)
-- Python implementation using standard library (zero dependencies)
-- Cross-platform compatible (Windows/macOS/Linux)
-- PEP 723 compliant (inline script metadata)
-- Runs with `--dangerously-bypass-approvals-and-sandbox` for automation (new sessions only)
+- **Binary distribution**: Single Go binary, zero dependencies
+- **Installation**: Download from GitHub Releases or use install.sh
+- **Cross-platform compatible**: Linux (amd64/arm64), macOS (amd64/arm64)
+- All automated runs must use the Bash tool with the fixed timeout to provide dual timeout protection and unified logging/exit semantics
+for automation (new sessions only)
 - Uses `--skip-git-repo-check` to work in any directory
 - Streams progress, returns only final agent message
 - Every execution returns a session ID for resuming conversations

skills/codex/scripts/codex.py

@@ -8,10 +8,14 @@ Codex CLI wrapper with cross-platform support and session management.
 **FIXED**: Auto-detect long inputs and use stdin mode to avoid shell argument issues.
 
 Usage:
-    New session:  uv run codex.py "task" [model] [workdir]
-    Resume:       uv run codex.py resume <session_id> "task" [model] [workdir]
+    New session:  uv run codex.py "task" [workdir]
+    Stdin mode:   uv run codex.py - [workdir]
+    Resume:       uv run codex.py resume <session_id> "task" [workdir]
+    Resume stdin: uv run codex.py resume <session_id> - [workdir]
     Alternative:  python3 codex.py "task"
     Direct exec:  ./codex.py "task"
+
+    Model configuration: Set CODEX_MODEL environment variable (default: gpt-5.1-codex)
 """
 import subprocess
 import json
@@ -19,11 +23,10 @@ import sys
 import os
 from typing import Optional
 
-DEFAULT_MODEL = 'gpt-5.1-codex'
+DEFAULT_MODEL = os.environ.get('CODEX_MODEL', 'gpt-5.1-codex')
 DEFAULT_WORKDIR = '.'
 DEFAULT_TIMEOUT = 7200  # 2 hours in seconds
 FORCE_KILL_DELAY = 5
-STDIN_THRESHOLD = 800  # Auto-switch to stdin for prompts longer than 800 chars
 
 
 def log_error(message: str):
@@ -36,6 +39,11 @@ def log_warn(message: str):
     sys.stderr.write(f"WARN: {message}\n")
 
 
+def log_info(message: str):
+    """输出信息到 stderr"""
+    sys.stderr.write(f"INFO: {message}\n")
+
+
 def resolve_timeout() -> int:
     """解析超时配置（秒）"""
     raw = os.environ.get('CODEX_TIMEOUT', '')
@@ -74,100 +82,132 @@ def parse_args():
         if len(sys.argv) < 4:
             log_error('Resume mode requires: resume <session_id> <task>')
             sys.exit(1)
+        task_arg = sys.argv[3]
         return {
             'mode': 'resume',
             'session_id': sys.argv[2],
-            'task': sys.argv[3],
-            'model': sys.argv[4] if len(sys.argv) > 4 else DEFAULT_MODEL,
-            'workdir': sys.argv[5] if len(sys.argv) > 5 else DEFAULT_WORKDIR
-        }
-    else:
-        return {
-            'mode': 'new',
-            'task': sys.argv[1],
-            'model': sys.argv[2] if len(sys.argv) > 2 else DEFAULT_MODEL,
-            'workdir': sys.argv[3] if len(sys.argv) > 3 else DEFAULT_WORKDIR
+            'task': task_arg,
+            'explicit_stdin': task_arg == '-',
+            'workdir': sys.argv[4] if len(sys.argv) > 4 else DEFAULT_WORKDIR,
         }
 
+    task_arg = sys.argv[1]
+    return {
+        'mode': 'new',
+        'task': task_arg,
+        'explicit_stdin': task_arg == '-',
+        'workdir': sys.argv[2] if len(sys.argv) > 2 else DEFAULT_WORKDIR,
+    }
 
-def build_codex_args(params: dict, use_stdin: bool) -> list:
+
+def read_piped_task() -> Optional[str]:
+    """
+    从 stdin 读取任务文本：
+    - 如果 stdin 是管道（非 tty）且存在内容，返回读取到的字符串
+    - 否则返回 None
+    """
+    stdin = sys.stdin
+    if stdin is None or stdin.isatty():
+        log_info("Stdin is tty or None, skipping pipe read")
+        return None
+    log_info("Reading from stdin pipe...")
+    data = stdin.read()
+    if not data:
+        log_info("Stdin pipe returned empty data")
+        return None
+
+    log_info(f"Read {len(data)} bytes from stdin pipe")
+    return data
+
+
+def should_stream_via_stdin(task_text: str, piped: bool) -> bool:
+    """
+    判定是否通过 stdin 传递任务：
+    - 有管道输入
+    - 文本包含换行
+    - 文本包含反斜杠
+    - 文本长度 > 800
+    """
+    if piped:
+        return True
+    if '\n' in task_text:
+        return True
+    if '\\' in task_text:
+        return True
+    if len(task_text) > 800:
+        return True
+    return False
+
+
+def build_codex_args(params: dict, target_arg: str) -> list:
     """
     构建 codex CLI 参数
 
     Args:
         params: 参数字典
-        use_stdin: 是否使用 stdin 模式（不在命令行参数中传递 task）
+        target_arg: 最终传递给 codex 的参数（'-' 或具体 task 文本）
     """
     if params['mode'] == 'resume':
-        if use_stdin:
-            return [
-                'codex', 'e',
-                '--skip-git-repo-check',
-                '--json',
-                'resume',
-                params['session_id'],
-                '-'  # 从 stdin 读取
-            ]
-        else:
-            return [
-                'codex', 'e',
-                '--skip-git-repo-check',
-                '--json',
-                'resume',
-                params['session_id'],
-                params['task']
-            ]
+        return [
+            'codex', 'e',
+            '-m', DEFAULT_MODEL,
+            '--skip-git-repo-check',
+            '--json',
+            'resume',
+            params['session_id'],
+            target_arg
+        ]
     else:
         base_args = [
             'codex', 'e',
-            '-m', params['model'],
+            '-m', DEFAULT_MODEL,
             '--dangerously-bypass-approvals-and-sandbox',
             '--skip-git-repo-check',
             '-C', params['workdir'],
-            '--json'
+            '--json',
+            target_arg
         ]
 
-        if use_stdin:
-            base_args.append('-')  # 从 stdin 读取
-        else:
-            base_args.append(params['task'])
-
         return base_args
 
 
-def main():
-    params = parse_args()
-    timeout_sec = resolve_timeout()
-
-    # **FIX: Auto-detect long inputs and enable stdin mode**
-    task_length = len(params['task'])
-    use_stdin = task_length > STDIN_THRESHOLD
-
-    if use_stdin:
-        log_warn(f"Task length ({task_length} chars) exceeds threshold, using stdin mode to avoid shell escaping issues")
-
-    codex_args = build_codex_args(params, use_stdin)
-
+def run_codex_process(codex_args, task_text: str, use_stdin: bool, timeout_sec: int):
+    """
+    启动 codex 子进程，处理 stdin / JSON 行输出和错误，成功时返回 (last_agent_message, thread_id)。
+    失败路径上负责日志和退出码。
+    """
     thread_id: Optional[str] = None
     last_agent_message: Optional[str] = None
+    process: Optional[subprocess.Popen] = None
 
     try:
-        # 启动 codex 子进程
+        # 启动 codex 子进程（文本模式管道）
+        log_info(f"Starting codex with args: {' '.join(codex_args[:5])}...")
         process = subprocess.Popen(
             codex_args,
-            stdin=subprocess.PIPE if use_stdin else None,  # **FIX: Enable stdin**
+            stdin=subprocess.PIPE if use_stdin else None,
             stdout=subprocess.PIPE,
-            stderr=sys.stderr,  # 错误直接透传到 stderr
+            stderr=sys.stderr,
             text=True,
-            bufsize=1  # 行缓冲
+            bufsize=1,
         )
+        log_info(f"Process started with PID: {process.pid}")
 
-        # **FIX: 如果使用 stdin 模式，写入任务到 stdin**
-        if use_stdin:
-            process.stdin.write(params['task'])
+        # 如果使用 stdin 模式，写入任务到 stdin 并关闭
+        if use_stdin and process.stdin is not None:
+            log_info(f"Writing {len(task_text)} chars to stdin...")
+            process.stdin.write(task_text)
+            process.stdin.flush()  # 强制刷新缓冲区，避免大任务死锁
             process.stdin.close()
+            log_info("Stdin closed")
 
         # 逐行解析 JSON 输出
+        if process.stdout is None:
+            log_error('Codex stdout pipe not available')
+            sys.exit(1)
+
+        log_info("Reading stdout...")
+
         for line in process.stdout:
             line = line.strip()
             if not line:
@@ -190,33 +230,26 @@ def main():
             except json.JSONDecodeError:
                 log_warn(f"Failed to parse line: {line}")
 
-        # 等待进程结束
+        # 等待进程结束并检查退出码
         returncode = process.wait(timeout=timeout_sec)
-
-        if returncode == 0:
-            if last_agent_message:
-                # 输出 agent_message
-                sys.stdout.write(f"{last_agent_message}\n")
-
-                # 输出 session_id（如果存在）
-                if thread_id:
-                    sys.stdout.write(f"\n---\nSESSION_ID: {thread_id}\n")
-
-                sys.exit(0)
-            else:
-                log_error('Codex completed without agent_message output')
-                sys.exit(1)
-        else:
+        if returncode != 0:
             log_error(f'Codex exited with status {returncode}')
             sys.exit(returncode)
 
+        if not last_agent_message:
+            log_error('Codex completed without agent_message output')
+            sys.exit(1)
+
+        return last_agent_message, thread_id
+
     except subprocess.TimeoutExpired:
         log_error('Codex execution timeout')
-        process.kill()
-        try:
-            process.wait(timeout=FORCE_KILL_DELAY)
-        except subprocess.TimeoutExpired:
-            pass
+        if process is not None:
+            process.kill()
+            try:
+                process.wait(timeout=FORCE_KILL_DELAY)
+            except subprocess.TimeoutExpired:
+                pass
         sys.exit(124)
 
     except FileNotFoundError:
@@ -224,13 +257,76 @@ def main():
         sys.exit(127)
 
     except KeyboardInterrupt:
-        process.terminate()
-        try:
-            process.wait(timeout=FORCE_KILL_DELAY)
-        except subprocess.TimeoutExpired:
-            process.kill()
+        log_error("Codex interrupted by user")
+        if process is not None:
+            process.terminate()
+            try:
+                process.wait(timeout=FORCE_KILL_DELAY)
+            except subprocess.TimeoutExpired:
+                process.kill()
         sys.exit(130)
 
 
+def main():
+    log_info("Script started")
+    params = parse_args()
+    log_info(f"Parsed args: mode={params['mode']}, task_len={len(params['task'])}")
+    timeout_sec = resolve_timeout()
+    log_info(f"Timeout: {timeout_sec}s")
+
+    explicit_stdin = params.get('explicit_stdin', False)
+
+    if explicit_stdin:
+        log_info("Explicit stdin mode: reading task from stdin")
+        task_text = sys.stdin.read()
+        if not task_text:
+            log_error("Explicit stdin mode requires task input from stdin")
+            sys.exit(1)
+        piped = not sys.stdin.isatty()
+    else:
+        piped_task = read_piped_task()
+        piped = piped_task is not None
+        task_text = piped_task if piped else params['task']
+
+    use_stdin = explicit_stdin or should_stream_via_stdin(task_text, piped)
+
+    if use_stdin:
+        reasons = []
+        if piped:
+            reasons.append('piped input')
+        if explicit_stdin:
+            reasons.append('explicit "-"')
+        if '\n' in task_text:
+            reasons.append('newline')
+        if '\\' in task_text:
+            reasons.append('backslash')
+        if len(task_text) > 800:
+            reasons.append('length>800')
+
+        if reasons:
+            log_warn(f"Using stdin mode for task due to: {', '.join(reasons)}")
+
+    target_arg = '-' if use_stdin else params['task']
+    codex_args = build_codex_args(params, target_arg)
+
+    log_info('codex running...')
+
+    last_agent_message, thread_id = run_codex_process(
+        codex_args=codex_args,
+        task_text=task_text,
+        use_stdin=use_stdin,
+        timeout_sec=timeout_sec,
+    )
+
+    # 输出 agent_message
+    sys.stdout.write(f"{last_agent_message}\n")
+
+    # 输出 session_id（如果存在）
+    if thread_id:
+        sys.stdout.write(f"\n---\nSESSION_ID: {thread_id}\n")
+
+    sys.exit(0)
+
+
 if __name__ == '__main__':
     main()

skills/gemini/SKILL.md

@@ -0,0 +1,120 @@
+---
+name: gemini
+description: Execute Gemini CLI for AI-powered code analysis and generation. Use when you need to leverage Google's Gemini models for complex reasoning tasks.
+---
+
+# Gemini CLI Integration
+
+## Overview
+
+Execute Gemini CLI commands with support for multiple models and flexible prompt input. Integrates Google's Gemini AI models into Claude Code workflows.
+
+## When to Use
+
+- Complex reasoning tasks requiring advanced AI capabilities
+- Code generation and analysis with Gemini models
+- Tasks requiring Google's latest AI technology
+- Alternative perspective on code problems
+
+## Usage
+**Mandatory**: Run via uv with fixed timeout 7200000ms (foreground):
+```bash
+uv run ~/.claude/skills/gemini/scripts/gemini.py "<prompt>" [working_dir]
+```
+
+**Optional** (direct execution or using Python):
+```bash
+~/.claude/skills/gemini/scripts/gemini.py "<prompt>" [working_dir]
+# or
+python3 ~/.claude/skills/gemini/scripts/gemini.py "<prompt>" [working_dir]
+```
+
+## Environment Variables
+
+- **GEMINI_MODEL**: Configure model (default: `gemini-3-pro-preview`)
+  - Example: `export GEMINI_MODEL=gemini-3`
+
+## Timeout Control
+
+- **Fixed**: 7200000 milliseconds (2 hours), immutable
+- **Bash tool**: Always set `timeout: 7200000` for double protection
+
+### Parameters
+
+- `prompt` (required): Task prompt or question
+- `working_dir` (optional): Working directory (default: current directory)
+
+### Return Format
+
+Plain text output from Gemini:
+
+```text
+Model response text here...
+```
+
+Error format (stderr):
+
+```text
+ERROR: Error message
+```
+
+### Invocation Pattern
+
+When calling via Bash tool, always include the timeout parameter:
+
+```yaml
+Bash tool parameters:
+- command: uv run ~/.claude/skills/gemini/scripts/gemini.py "<prompt>"
+- timeout: 7200000
+- description: <brief description of the task>
+```
+
+Alternatives:
+
+```yaml
+# Direct execution (simplest)
+- command: ~/.claude/skills/gemini/scripts/gemini.py "<prompt>"
+
+# Using python3
+- command: python3 ~/.claude/skills/gemini/scripts/gemini.py "<prompt>"
+```
+
+### Examples
+
+**Basic query:**
+
+```bash
+uv run ~/.claude/skills/gemini/scripts/gemini.py "explain quantum computing"
+# timeout: 7200000
+```
+
+**Code analysis:**
+
+```bash
+uv run ~/.claude/skills/gemini/scripts/gemini.py "review this code for security issues: $(cat app.py)"
+# timeout: 7200000
+```
+
+**With specific working directory:**
+
+```bash
+uv run ~/.claude/skills/gemini/scripts/gemini.py "analyze project structure" "/path/to/project"
+# timeout: 7200000
+```
+
+**Using python3 directly (alternative):**
+
+```bash
+python3 ~/.claude/skills/gemini/scripts/gemini.py "your prompt here"
+```
+
+## Notes
+
+- **Recommended**: Use `uv run` for automatic Python environment management (requires uv installed)
+- **Alternative**: Direct execution `./gemini.py` (uses system Python via shebang)
+- Python implementation using standard library (zero dependencies)
+- Cross-platform compatible (Windows/macOS/Linux)
+- PEP 723 compliant (inline script metadata)
+- Requires Gemini CLI installed and authenticated
+- Supports all Gemini model variants (configure via `GEMINI_MODEL` environment variable)
+- Output is streamed directly from Gemini CLI

skills/gemini/scripts/gemini.py

@@ -0,0 +1,140 @@
+#!/usr/bin/env python3
+# /// script
+# requires-python = ">=3.8"
+# dependencies = []
+# ///
+"""
+Gemini CLI wrapper with cross-platform support.
+
+Usage:
+    uv run gemini.py "<prompt>" [workdir]
+    python3 gemini.py "<prompt>"
+    ./gemini.py "your prompt"
+"""
+import subprocess
+import sys
+import os
+
+DEFAULT_MODEL = os.environ.get('GEMINI_MODEL', 'gemini-3-pro-preview')
+DEFAULT_WORKDIR = '.'
+TIMEOUT_MS = 7_200_000  # 固定 2 小时，毫秒
+DEFAULT_TIMEOUT = TIMEOUT_MS // 1000
+FORCE_KILL_DELAY = 5
+
+
+def log_error(message: str):
+    """输出错误信息到 stderr"""
+    sys.stderr.write(f"ERROR: {message}\n")
+
+
+def log_warn(message: str):
+    """输出警告信息到 stderr"""
+    sys.stderr.write(f"WARN: {message}\n")
+
+
+def log_info(message: str):
+    """输出信息到 stderr"""
+    sys.stderr.write(f"INFO: {message}\n")
+
+
+def parse_args():
+    """解析位置参数"""
+    if len(sys.argv) < 2:
+        log_error('Prompt required')
+        sys.exit(1)
+
+    return {
+        'prompt': sys.argv[1],
+        'workdir': sys.argv[2] if len(sys.argv) > 2 else DEFAULT_WORKDIR
+    }
+
+
+def build_gemini_args(args) -> list:
+    """构建 gemini CLI 参数"""
+    return [
+        'gemini',
+        '-m', DEFAULT_MODEL,
+        '-p', args['prompt']
+    ]
+
+
+def main():
+    log_info('Script started')
+    args = parse_args()
+    log_info(f"Prompt length: {len(args['prompt'])}")
+    log_info(f"Working dir: {args['workdir']}")
+    gemini_args = build_gemini_args(args)
+    timeout_sec = DEFAULT_TIMEOUT
+    log_info(f"Timeout: {timeout_sec}s")
+
+    # 如果指定了工作目录，切换到该目录
+    if args['workdir'] != DEFAULT_WORKDIR:
+        try:
+            os.chdir(args['workdir'])
+        except FileNotFoundError:
+            log_error(f"Working directory not found: {args['workdir']}")
+            sys.exit(1)
+        except PermissionError:
+            log_error(f"Permission denied: {args['workdir']}")
+            sys.exit(1)
+        log_info('Changed working directory')
+
+    try:
+        log_info(f"Starting gemini with model {DEFAULT_MODEL}")
+        process = None
+        # 启动 gemini 子进程，直接透传 stdout 和 stderr
+        process = subprocess.Popen(
+            gemini_args,
+            stdout=subprocess.PIPE,
+            stderr=subprocess.PIPE,
+            text=True,
+            bufsize=1  # 行缓冲
+        )
+
+        # 实时输出 stdout
+        for line in process.stdout:
+            sys.stdout.write(line)
+            sys.stdout.flush()
+
+        # 等待进程结束
+        returncode = process.wait(timeout=timeout_sec)
+
+        # 读取 stderr
+        stderr_output = process.stderr.read()
+        if stderr_output:
+            sys.stderr.write(stderr_output)
+
+        # 检查退出码
+        if returncode != 0:
+            log_error(f'Gemini exited with status {returncode}')
+            sys.exit(returncode)
+
+        sys.exit(0)
+
+    except subprocess.TimeoutExpired:
+        log_error(f'Gemini execution timeout ({timeout_sec}s)')
+        if process is not None:
+            process.kill()
+            try:
+                process.wait(timeout=FORCE_KILL_DELAY)
+            except subprocess.TimeoutExpired:
+                pass
+        sys.exit(124)
+
+    except FileNotFoundError:
+        log_error("gemini command not found in PATH")
+        log_error("Please install Gemini CLI: https://github.com/google/generative-ai-python")
+        sys.exit(127)
+
+    except KeyboardInterrupt:
+        if process is not None:
+            process.terminate()
+            try:
+                process.wait(timeout=FORCE_KILL_DELAY)
+            except subprocess.TimeoutExpired:
+                process.kill()
+        sys.exit(130)
+
+
+if __name__ == '__main__':
+    main()