tackle-harness 0.0.2

package/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 tackle contributors
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

package/README.en.md

@@ -0,0 +1,259 @@
+# Tackle Harness
+
+> A plugin-based AI Agent workflow framework that provides task management, workflow orchestration, and role management for Claude Code
+
+[![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](https://opensource.org/licenses/MIT)
+[![Version](https://img.shields.io/badge/version-0.0.2-blue.svg)](https://github.com/ph419/tackle)
+
+**[中文文档](README.md)**
+
+## Why Tackle Harness
+
+You describe what you need; Tackle Harness manages the entire process:
+
+- **Plan first, human approval required** — AI outputs implementation plans and work package breakdowns, then waits for your confirmation before writing any code. No more "AI went rogue and changed a bunch of things."
+- **Complex requirements, parallel delivery** — Large requirements are automatically split into independent modules, with multiple agents working simultaneously. Frontend, backend, and database changes progress in parallel — no serial waiting.
+- **Experience accumulation, gets better over time** — After each task, lessons learned are automatically extracted. Next time a similar issue arises, agents reference historical experience for better decisions.
+
+### End-to-End Data Flow
+
+User requirements pass through five stages from planning to delivery:
+
+```mermaid
+flowchart LR
+    REQ["User Requirement<br/>(Natural Language)"]
+    P0["P0: Plan<br/>task-creator / split-wp"]
+    P1["P1: Review<br/>human-checkpoint"]
+    P2["P2: Execute<br/>agent-dispatcher"]
+    P3["P3: Verify<br/>checklist / experience-logger"]
+    P4["P4: Report<br/>completion-report"]
+
+    O1["docs/wp/*.md<br/>task.md updated"]
+    O2["User Approval<br/>(Human Intervention)"]
+    O3["Agent Teams<br/>(Multi-agent Parallel)"]
+    O4["Completion Report<br/>Experience Logged"]
+
+    REQ --> P0 --> P1 --> P2 --> P3 --> P4
+
+    P0 -.-> O1
+    P1 -.-> O2
+    P2 -.-> O3
+    P3 -.-> O4
+```
+
+## Installation
+
+```bash
+npm install tackle-harness
+```
+
+## Quick Start
+
+```bash
+# Navigate to your project directory
+cd your-project
+
+# One-command initialization (build skills + register hooks + create config directories)
+npx tackle-harness init
+
+# Or step by step
+npx tackle-harness build      # Build skills to .claude/skills/, merge hooks into settings.json
+npx tackle-harness validate   # Validate plugin integrity
+```
+
+## Use Cases
+
+### Case 1: New Feature Development
+
+**Your situation**: Adding a "Team Collaboration" module to a SaaS product, involving frontend UI, backend API, and database changes.
+
+**Just say**:
+```
+Start workflow, implement team collaboration module, including:
+- Team creation and management pages
+- Member invitation and permission APIs
+- Database table design
+```
+
+**What Tackle Harness does**:
+1. Analyzes requirement complexity, splits into 4 work packages (frontend, backend, database, integration testing)
+2. Outputs implementation plans for each work package, pauses for your review
+3. After your approval, dispatches multiple agents to develop modules in parallel
+4. Automatically runs code checks and test verification
+5. Generates a completion report and asks for next steps
+
+**Skills involved**: workflow-orchestrator → split-work-package → human-checkpoint → agent-dispatcher → checklist → completion-report
+
+### Case 2: Batch Bug Fixes
+
+**Your situation**: A backlog of 5 bugs before the sprint ends, want to handle them in parallel and wrap up quickly.
+
+**Just say**:
+```
+Batch execute WP-015 through WP-019, fix these 5 bugs in parallel
+```
+
+**What Tackle Harness does**:
+1. Analyzes dependencies between the 5 bugs (check for overlapping file changes)
+2. Assigns conflict-free bugs to different agents for simultaneous fixes
+3. Queues dependent bugs sequentially, auto-starting the next after each completes
+4. Runs a checklist after all fixes to confirm no new issues introduced
+
+**Skills involved**: agent-dispatcher → checklist → completion-report
+
+### Case 3: System Refactoring
+
+**Your situation**: Need to refactor a monolithic application into microservices, involving coordinated changes across multiple modules, worried about breaking things.
+
+**Just say**:
+```
+Split work package, extract the user module from the monolith into an independent service
+```
+
+**What Tackle Harness does**:
+1. Deep analysis of code structure, identifies all modules and dependencies that need changes
+2. Generates a detailed refactoring plan (interface extraction, data migration, routing adjustments, etc.)
+3. Pauses for your review of the architecture proposal (critical decision point)
+4. Executes refactoring in dependency-ordered batches, auto-verifying after each batch
+5. Records refactoring experience for future similar extraction tasks
+
+**Skills involved**: split-work-package → human-checkpoint → agent-dispatcher → checklist → experience-logger → completion-report
+
+## Command Reference
+
+| Command | Description |
+|---------|-------------|
+| `npx tackle-harness` | Default: runs build |
+| `npx tackle-harness build` | Build all skills, update .claude/settings.json |
+| `npx tackle-harness validate` | Validate plugin format |
+| `npx tackle-harness init` | First-time setup: build + create .claude/ directories |
+| `npx tackle-harness --root <path>` | Specify target project path (default: current directory) |
+| `npx tackle-harness --help` | Show help |
+
+## Skills Reference
+
+| Skill | Trigger (CN / EN) | Function |
+|-------|---------|----------|
+| task-creator | "创建任务" / "create task" | Create a single task in the task list |
+| batch-task-creator | "批量创建任务" / "batch create tasks" | Batch create multiple tasks |
+| split-work-package | "拆分工作包" / "split work package" | Split requirements into executable work packages |
+| progress-tracker | "记录进度" / "record progress" | Track and report work progress |
+| team-cleanup | "清理团队" / "cleanup team" | Release residual team resources |
+| human-checkpoint | "等待审核" / "wait for review" | Pause and request human confirmation |
+| role-manager | "查看角色" / "view roles" | Manage project role definitions |
+| checklist | "运行检查" / "run checklist" | Execute checklists |
+| completion-report | "完成报告" / "completion report" | Generate completion report |
+| experience-logger | "总结经验" / "log experience" | Record project lessons learned |
+| agent-dispatcher | "批量执行" / "dispatch agents" | Dispatch multiple sub-agents in parallel |
+| workflow-orchestrator | "开始工作流" / "start workflow" | Orchestrate complete workflows |
+
+## Workflow Overview
+
+User requirements pass through 5 stages from planning to delivery:
+
+```
+Requirement → Plan(P0) → Review(P1) → Execute(P2) → Verify(P3) → Report(P4) → Delivery
+```
+
+| Stage | What Happens | Key Skills |
+|-------|-------------|------------|
+| **P0 Plan** | Parse requirements, split into work packages, write docs | task-creator, split-work-package |
+| **P1 Review** | Pause for your plan approval (mandatory human intervention) | human-checkpoint |
+| **P2 Execute** | Multi-agent parallel development, scheduled by dependencies | agent-dispatcher |
+| **P3 Verify** | Code/test/doc quality verification, extract experience | checklist, experience-logger |
+| **P4 Report** | Generate completion report, ask for next steps | completion-report |
+
+> For the full data flow diagram and stage details, see [docs/ai_workflow.md](docs/ai_workflow.md)
+
+## Plugin Architecture
+
+Tackle Harness contains 4 plugin types, 18 plugins total:
+
+| Type | Count | Purpose |
+|------|-------|---------|
+| Skill | 12 | Executable skills, directly callable by Claude Code |
+| Provider | 3 | State store, role registry, memory store |
+| Hook | 1 | Skill gate, intercepts edit operations and skill calls |
+| Validator | 2 | Document sync validation, work package validation |
+
+> For plugin dependency graph and development guide, see [docs/plugin-development.md](docs/plugin-development.md)
+
+## Build Output Structure
+
+After running `tackle-harness build`, the following is generated in your project:
+
+```
+your-project/
+  .claude/
+    skills/                          # 12 skills
+      skill-task-creator/skill.md
+      skill-batch-task-creator/skill.md
+      skill-split-work-package/skill.md
+      skill-progress-tracker/skill.md
+      skill-team-cleanup/skill.md
+      skill-human-checkpoint/skill.md
+      skill-role-manager/skill.md
+      skill-checklist/skill.md
+      skill-completion-report/skill.md
+      skill-experience-logger/skill.md
+      skill-agent-dispatcher/skill.md
+      skill-workflow-orchestrator/skill.md
+    settings.json                    # Auto-registered hooks
+```
+
+## FAQ
+
+### Skills not working after installation?
+
+Make sure you ran `npx tackle-harness build` in your project root, and that `.claude/skills/` contains 12 skill folders.
+
+### Can multiple projects share an installation?
+
+Each project installs and builds independently. Different projects can use different versions.
+
+### Global installation
+
+```bash
+npm install -g tackle-harness
+tackle-harness build
+```
+
+After global installation, use the `tackle-harness` command directly without `npx`.
+
+### How to uninstall?
+
+```bash
+npm uninstall tackle-harness
+```
+
+Skill files remain in `.claude/skills/`. Delete manually if needed.
+
+### What are the hooks in settings.json?
+
+`tackle-harness build` automatically injects two hooks into `.claude/settings.json`:
+- `PreToolUse(Edit|Write)` — Blocks file edits under certain states
+- `PostToolUse(Skill)` — Updates state after skill calls
+
+These hooks point to scripts in `node_modules/tackle-harness/` and won't affect other configurations in your project. Existing settings.json content is preserved; only tackle-harness-related hooks are appended.
+
+## Documentation
+
+- [Configuration Reference](docs/config-reference.md) - Complete configuration file documentation
+- [Best Practices](docs/best-practices.md) - Usage tips and optimization techniques
+- [Plugin Development](docs/plugin-development.md) - Plugin architecture and development guide
+- [Workflow Details](docs/ai_workflow.md) - Full workflow data flow and stage descriptions
+
+## Contributing
+
+Contributions welcome! We accept bug reports, feature suggestions, code submissions, and documentation improvements. See [Contributing Guide](CONTRIBUTING.md).
+
+Quick start: Fork → Create branch → Make changes → Submit PR. Follow [Conventional Commits](https://www.conventionalcommits.org/) format.
+
+## License
+
+MIT License - See [LICENSE](LICENSE) file
+
+## Acknowledgments
+
+This project draws on excellent designs from the following open-source projects:
+- DeerFlow - Memory extraction and middleware architecture

package/README.md

@@ -0,0 +1,261 @@
+# Tackle Harness
+
+> 基于插件的 AI Agent 工作流框架，为 Claude Code 提供任务管理、工作流编排、角色管理等能力
+
+[![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](https://opensource.org/licenses/MIT)
+[![Version](https://img.shields.io/badge/version-0.0.2-blue.svg)](https://github.com/ph419/tackle)
+
+**[English](README.en.md)**
+
+## 为什么选择 Tackle Harness
+
+你告诉 AI 需求，Tackle Harness 帮你管好整个流程：
+
+- **方案先行，人工把关** — AI 先输出实施方案和工作包拆分，等你确认后才动手写代码。不会出现「AI 自作主张改了一堆东西」的情况。
+- **复杂需求，并行交付** — 大需求自动拆成多个独立模块，调度多个 Agent 同时工作。前后端、数据库变更同步推进，不用串行等待。
+- **经验沉淀，越用越好** — 每次任务完成后自动提炼经验教训。下次遇到类似问题时，Agent 会参考历史经验做出更好的决策。
+
+
+
+### 端到端数据流
+
+用户需求经五个阶段完成从规划到交付的完整生命周期：
+
+```mermaid
+flowchart LR
+    REQ["用户需求<br/>(自然语言)"]
+    P0["P0: 规划<br/>task-creator / split-wp"]
+    P1["P1: 审核<br/>human-checkpoint"]
+    P2["P2: 执行<br/>agent-dispatcher"]
+    P3["P3: 检查<br/>checklist / experience-logger"]
+    P4["P4: 汇报<br/>completion-report"]
+
+    O1["docs/wp/*.md<br/>task.md 更新"]
+    O2["用户确认/修改<br/>(人工介入)"]
+    O3["Agent Teams<br/>(多代理并行)"]
+    O4["完成报告<br/>经验沉淀"]
+
+    REQ --> P0 --> P1 --> P2 --> P3 --> P4
+
+    P0 -.-> O1
+    P1 -.-> O2
+    P2 -.-> O3
+    P3 -.-> O4
+```
+
+## 安装
+
+```bash
+npm install tackle-harness
+```
+
+## 快速开始
+
+```bash
+# 进入你的项目目录
+cd your-project
+
+# 一键初始化（构建技能 + 注册钩子 + 创建配置目录）
+npx tackle-harness init
+
+# 或者分步执行
+npx tackle-harness build      # 构建技能到 .claude/skills/，合并 hooks 到 settings.json
+npx tackle-harness validate   # 验证插件完整性
+```
+
+## 使用场景
+
+### 场景一：新功能开发
+
+**你的情况**：需要为 SaaS 产品添加「团队协作」模块，涉及前端界面、后端 API 和数据库变更。
+
+**你只需要说**：
+```
+开始工作流，实现团队协作模块，包括：
+- 团队创建和管理页面
+- 成员邀请和权限 API
+- 数据库表设计
+```
+
+**Tackle Harness 会做什么**：
+1. 分析需求复杂度，拆分为 4 个工作包（前端、后端、数据库、集成测试）
+2. 输出每个工作包的实施方案，暂停等你审核
+3. 你确认后，调度多个 Agent 并行开发各模块
+4. 自动执行代码检查和测试验证
+5. 生成完成报告，询问你下一步
+
+**涉及技能**：workflow-orchestrator → split-work-package → human-checkpoint → agent-dispatcher → checklist → completion-report
+
+### 场景二：Bug 批量修复
+
+**你的情况**：Sprint 结束前积压了 5 个 Bug，希望能并行处理尽快收尾。
+
+**你只需要说**：
+```
+批量执行 WP-015 到 WP-019，并行修复这 5 个 Bug
+```
+
+**Tackle Harness 会做什么**：
+1. 分析 5 个 Bug 之间的依赖关系（有没有改动同一文件）
+2. 无冲突的 Bug 分配给不同 Agent 同时修复
+3. 有依赖的 Bug 按顺序排队，前一个完成后自动启动下一个
+4. 全部修复后运行检查清单，确认没有引入新问题
+
+**涉及技能**：agent-dispatcher → checklist → completion-report
+
+### 场景三：系统重构
+
+**你的情况**：需要将单体应用拆分为微服务架构，涉及多个模块的协调改动，担心改出问题。
+
+**你只需要说**：
+```
+拆分工作包，将用户模块从单体应用中拆分为独立服务
+```
+
+**Tackle Harness 会做什么**：
+1. 深入分析代码结构，识别所有需要改动的模块和依赖关系
+2. 生成详细的重构计划（接口抽取、数据迁移、路由调整等）
+3. 暂停等你审核架构方案（这是关键决策点）
+4. 按依赖顺序分批执行重构，每批完成后自动验证
+5. 记录重构经验，下次类似的拆分任务可以直接参考
+
+**涉及技能**：split-work-package → human-checkpoint → agent-dispatcher → checklist → experience-logger → completion-report
+
+## 命令一览
+
+| 命令 | 说明 |
+|------|------|
+| `npx tackle-harness` | 默认执行 build |
+| `npx tackle-harness build` | 构建所有技能，更新 .claude/settings.json |
+| `npx tackle-harness validate` | 验证插件格式是否正确 |
+| `npx tackle-harness init` | 首次安装：build + 创建 .claude/ 目录 |
+| `npx tackle-harness --root <path>` | 指定目标项目路径（默认为当前目录） |
+| `npx tackle-harness --help` | 查看帮助信息 |
+
+## 技能清单
+
+| 技能 | 触发方式 | 功能 |
+|------|----------|------|
+| task-creator | "创建任务" / "create task" | 创建单个任务到任务列表 |
+| batch-task-creator | "批量创建任务" / "batch create tasks" | 批量创建多个任务 |
+| split-work-package | "拆分工作包" / "split work package" | 将需求拆分为可执行的工作包 |
+| progress-tracker | "记录进度" / "record progress" | 追踪和汇报工作进度 |
+| team-cleanup | "清理团队" / "cleanup team" | 释放残留的团队资源 |
+| human-checkpoint | "等待审核" / "wait for review" | 暂停并请求人工确认 |
+| role-manager | "查看角色" / "view roles" | 管理项目角色定义 |
+| checklist | "运行检查" / "run checklist" | 执行检查清单 |
+| completion-report | "完成报告" / "completion report" | 生成完成报告 |
+| experience-logger | "总结经验" / "log experience" | 记录项目经验教训 |
+| agent-dispatcher | "批量执行" / "dispatch agents" | 调度多个子代理并行工作 |
+| workflow-orchestrator | "开始工作流" / "start workflow" | 编排完整工作流 |
+
+## 工作流概览
+
+用户需求经过 5 个阶段完成从规划到交付：
+
+```
+需求 → 规划(P0) → 审核(P1) → 执行(P2) → 检查(P3) → 汇报(P4) → 交付
+```
+
+| 阶段 | 做什么 | 关键技能 |
+|------|--------|----------|
+| **P0 规划** | 解析需求，拆分为工作包，写入文档 | task-creator, split-work-package |
+| **P1 审核** | 暂停等待你确认方案（强制人工介入） | human-checkpoint |
+| **P2 执行** | 多 Agent 并行开发，按依赖调度 | agent-dispatcher |
+| **P3 检查** | 代码/测试/文档质量验证，提炼经验 | checklist, experience-logger |
+| **P4 汇报** | 生成完成报告，询问下一步 | completion-report |
+
+> 完整的数据流图和阶段细节请参阅 [docs/ai_workflow.md](docs/ai_workflow.md)
+
+## 插件架构
+
+Tackle Harness 包含四类插件，共 18 个：
+
+| 类型 | 数量 | 作用 |
+|------|------|------|
+| Skill | 12 | 可执行技能，Claude Code 直接调用 |
+| Provider | 3 | 状态存储、角色注册、记忆存储 |
+| Hook | 1 | 技能门控，拦截编辑操作和技能调用 |
+| Validator | 2 | 文档同步验证、工作包验证 |
+
+> 插件依赖关系和开发指南请参阅 [docs/plugin-development.md](docs/plugin-development.md)
+
+## 构建后的项目结构
+
+执行 `tackle-harness build` 后，你的项目中会生成以下内容：
+
+```
+your-project/
+  .claude/
+    skills/                          # 12 个技能
+      skill-task-creator/skill.md
+      skill-batch-task-creator/skill.md
+      skill-split-work-package/skill.md
+      skill-progress-tracker/skill.md
+      skill-team-cleanup/skill.md
+      skill-human-checkpoint/skill.md
+      skill-role-manager/skill.md
+      skill-checklist/skill.md
+      skill-completion-report/skill.md
+      skill-experience-logger/skill.md
+      skill-agent-dispatcher/skill.md
+      skill-workflow-orchestrator/skill.md
+    settings.json                    # 自动注册的 hooks
+```
+
+## 常见问题
+
+### 安装后技能没有生效？
+
+确保在项目根目录执行了 `npx tackle-harness build`，并且 `.claude/skills/` 目录下有 12 个技能文件夹。
+
+### 多个项目能否共用？
+
+每个项目独立安装、独立构建。不同项目可以安装不同版本。
+
+### 全局安装
+
+```bash
+npm install -g tackle-harness
+tackle-harness build
+```
+
+全局安装后直接使用 `tackle-harness` 命令，无需 `npx`。
+
+### 如何卸载？
+
+```bash
+npm uninstall tackle-harness
+```
+
+技能文件会保留在 `.claude/skills/` 中，如需清理请手动删除。
+
+### settings.json 中的 hooks 是什么？
+
+`tackle-harness build` 会自动向 `.claude/settings.json` 注入两个 hook：
+- `PreToolUse(Edit|Write)` — 在特定状态下阻止文件编辑
+- `PostToolUse(Skill)` — 技能调用后更新状态
+
+这些 hook 指向 `node_modules/tackle-harness/` 中的脚本，不会影响你项目中的其他配置。已有的 settings.json 内容会被保留，仅追加 tackle-harness 相关的 hooks。
+
+## 文档
+
+- [配置参考](docs/config-reference.md) - 完整的配置文件说明
+- [最佳实践](docs/best-practices.md) - 使用建议和优化技巧
+- [插件开发](docs/plugin-development.md) - 插件架构和开发指南
+- [工作流详解](docs/ai_workflow.md) - 完整的工作流数据流和阶段说明
+
+## 贡献
+
+欢迎贡献！我们接受 Bug 报告、功能建议、代码提交和文档改进。详见 [贡献指南](CONTRIBUTING.md)。
+
+快速上手：Fork → 创建分支 → 修改 → 提交 PR。Commit 遵循 [Conventional Commits](https://www.conventionalcommits.org/) 格式。
+
+## 许可证
+
+MIT License - 详见 [LICENSE](LICENSE) 文件
+
+## 致谢
+
+本项目借鉴了以下开源项目的优秀设计：
+- DeerFlow - 记忆提取和中间件架构

package/bin/tackle.js

@@ -0,0 +1,150 @@
+#!/usr/bin/env node
+/**
+ * tackle-harness - CLI entry point for the AI Agent Harness framework
+ *
+ * Usage:
+ *   tackle-harness             Build all plugins into .claude/skills/ and update settings.json
+ *   tackle-harness build       Same as above (default command)
+ *   tackle-harness validate    Validate plugin.json files without building
+ *   tackle-harness init        First-time setup: build + generate default config
+ *   tackle-harness --help      Show usage info
+ */
+
+'use strict';
+
+var path = require('path');
+var fs = require('fs');
+var HarnessBuild = require('../plugins/runtime/harness-build');
+
+// ---------------------------------------------------------------------------
+// Path resolution
+// ---------------------------------------------------------------------------
+
+/** Root directory of this npm package (where plugins/ lives) */
+var packageRoot = path.resolve(__dirname, '..');
+
+/** Target project root directory (where output goes) */
+var targetRoot = process.cwd();
+
+// ---------------------------------------------------------------------------
+// CLI arg parsing
+// ---------------------------------------------------------------------------
+
+var args = process.argv.slice(2);
+var command = args[0] || 'build';
+var flags = {
+  root: null,
+};
+
+for (var i = 0; i < args.length; i++) {
+  if (args[i] === '--root' && args[i + 1]) {
+    flags.root = args[++i];
+  }
+}
+
+// Override target root if --root flag provided
+if (flags.root) {
+  targetRoot = path.resolve(flags.root);
+}
+
+// Subcommand aliases
+if (command === '--validate') command = 'validate';
+if (command === '--help' || command === '-h') command = 'help';
+
+// ---------------------------------------------------------------------------
+// Helper: create builder instance with correct paths
+// ---------------------------------------------------------------------------
+
+function createBuilder() {
+  return new HarnessBuild({
+    rootDir: targetRoot,
+    registryPath: path.join(packageRoot, 'plugins', 'plugin-registry.json'),
+    pluginsDir: path.join(packageRoot, 'plugins', 'core'),
+    outputSkillsDir: path.join(targetRoot, '.claude', 'skills'),
+    outputHooksDir: path.join(targetRoot, '.claude', 'hooks'),
+  });
+}
+
+// ---------------------------------------------------------------------------
+// Commands
+// ---------------------------------------------------------------------------
+
+function cmdBuild() {
+  var builder = createBuilder();
+  var result = builder.build();
+
+  if (result.success) {
+    builder.updateSettings(targetRoot, packageRoot);
+  }
+
+  console.log(result.summary);
+
+  if (result.success) {
+    console.log('[tackle-harness] Settings updated: .claude/settings.json');
+    console.log('[tackle-harness] Done! Skills are ready to use.');
+  }
+
+  process.exit(result.success ? 0 : 1);
+}
+
+function cmdValidate() {
+  var builder = createBuilder();
+  var result = builder.validate();
+  console.log(result.summary);
+  process.exit(result.valid ? 0 : 1);
+}
+
+function cmdInit() {
+  console.log('[tackle-harness] Initializing...');
+  console.log('[tackle-harness] Target project: ' + targetRoot);
+  console.log('[tackle-harness] Package root:   ' + packageRoot);
+  console.log('');
+
+  // 1. Ensure .claude/ directory exists
+  var claudeDir = path.join(targetRoot, '.claude');
+  if (!fs.existsSync(claudeDir)) {
+    fs.mkdirSync(claudeDir, { recursive: true });
+    console.log('[tackle-harness] Created .claude/ directory');
+  }
+
+  // 2. Run build
+  cmdBuild();
+}
+
+function cmdHelp() {
+  console.log('tackle-harness - Plugin-based AI Agent Harness for Claude Code');
+  console.log('');
+  console.log('Usage:');
+  console.log('  tackle-harness             Build all plugins (default)');
+  console.log('  tackle-harness build       Build all plugins');
+  console.log('  tackle-harness validate    Validate plugin.json files');
+  console.log('  tackle-harness init        First-time setup (build + config)');
+  console.log('');
+  console.log('Options:');
+  console.log('  --root <path>      Specify target project root (default: cwd)');
+  console.log('  --help, -h         Show this help message');
+  console.log('');
+  console.log('After running tackle-harness build, skills are available in .claude/skills/');
+  console.log('and hooks are registered in .claude/settings.json');
+  process.exit(0);
+}
+
+// ---------------------------------------------------------------------------
+// Dispatch
+// ---------------------------------------------------------------------------
+
+switch (command) {
+  case 'build':
+    cmdBuild();
+    break;
+  case 'validate':
+    cmdValidate();
+    break;
+  case 'init':
+    cmdInit();
+    break;
+  case 'help':
+  default:
+    cmdHelp();
+    break;
+}