fix: skills/MCP/completion system overhaul — separate skill kinds, fix MCP remote servers, deduplicate completion processing

Skills system:
- Replace `CommandBadge.isSkill: boolean` with `kind: SkillKind` (agent_skill | slash_command | sdk_command | codepilot_command)
- handleSubmit dispatches by kind: agent_skill retains skill name in prompt, slash/sdk commands pass through to SDK, codepilot commands expand via COMMAND_PROMPTS
- Popover groups commands into Commands / Slash Commands / Agent Skills
- API `/api/skills` returns `kind` field; `displayOverride` persisted through full chain (frontend → API → DB)
- Forward SDK `system:init` slash_commands/skills metadata to frontend via SSE

MCP management:
- Fix remote server (SSE/HTTP) creation: validate `url` instead of `command`
- Fix dual-config read/write asymmetry: PUT routes servers by `_source` to correct file (~/.claude.json vs ~/.claude/settings.json)
- DELETE checks both config files; POST checks both for name collision
- McpManager preserves `_source` tags on edit/rename; JSON editor only manages settings.json servers

Completion processing:
- Extract core logic into `onboarding-processor.ts` and `checkin-processor.ts` (shared by route handlers and server-side detection)
- Remove frontend `detectAssistantCompletion` from ChatView — server-side path in `finally` block is now the single source of truth
- Replace hard-coded `fetch("http://localhost:PORT/...")` with direct function imports
- Both processors are internally idempotent (check state before processing)

Other:
- `settingSources` always includes 'user' so SDK loads ~/.claude/skills/
- Add 24 unit tests for SkillKind dispatch logic
- Add onboarding-completion extraction tests

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: 7418

docs/exec-plans/completed/skills-mcp-tooling-fix.md

@@ -0,0 +1,103 @@
+# Skills / MCP / Completion 系统修复
+
+**状态：** 已完成
+**日期：** 2026-03-10
+**关联调研：** `docs/research/skills-agent-sdk-review-2026-03-10.md`
+
+## 问题概述
+
+三个子系统存在多处 bug，部分为 P0（功能不可用）级别：
+
+### 1. Skills 系统：Agent Skills 与 Slash Commands 混淆
+
+**根因：** `CommandBadge.isSkill: boolean` 只区分"内置 vs 非内置"，无法区分 agent_skill / slash_command / sdk_command / codepilot_command 四种执行语义。
+
+**症状：**
+- Agent Skill 的 SKILL.md 被展开为用户消息发送，绕过 SDK 原生加载
+- 用户补充上下文后 `/command` 本体丢失
+- Popover 中无法区分不同类型的命令
+
+### 2. MCP 管理：远程服务器创建失败 + 双文件读写不一致
+
+**根因：**
+- POST 校验要求 `command` 字段，但 SSE/HTTP 服务器用 `url` 字段
+- GET 合并读取两个配置文件但 PUT 只写 settings.json
+- DELETE 只检查 settings.json
+
+### 3. Completion 处理：前端+后端双重执行 + 硬编码 localhost
+
+**根因：**
+- ChatView.tsx 和 route.ts 都处理 onboarding/checkin completion
+- 后端通过 `fetch("http://localhost:${PORT}")` 回调自身 API
+
+## 修改清单
+
+### Phase 1: SkillKind 类型系统
+
+| 文件 | 改动 |
+|------|------|
+| `src/types/index.ts` | 新增 `SkillKind` 类型；`MCPServerConfig.command` 改为可选 |
+| `src/app/api/skills/route.ts` | `SkillFile` 增加 `kind` 字段，各扫描函数返回正确 kind |
+| `src/app/api/skills/[name]/route.ts` | GET/PUT 响应包含 `kind` |
+
+### Phase 2: MessageInput 核心修复
+
+| 文件 | 改动 |
+|------|------|
+| `src/components/chat/MessageInput.tsx` | `isSkill` → `kind: SkillKind`；`handleSubmit` 按 kind 分流；Popover 分三组 |
+
+分流逻辑：
+- `agent_skill`: 发送 `Use the {name} skill. User context: {text}`，displayOverride 显示 `/name`
+- `slash_command` / `sdk_command`: 发送 `/{command} {text}` 原样给 SDK
+- `codepilot_command`: 展开 COMMAND_PROMPTS，displayOverride 显示 `/command`
+
+### Phase 3: displayOverride 持久化链路
+
+| 文件 | 改动 |
+|------|------|
+| `src/app/chat/page.tsx` | `sendFirstMessage` 接受 displayOverride 参数 |
+| `src/app/api/chat/route.ts` | 提取 displayOverride，DB 存储用 displayOverride |
+| `src/components/chat/ChatView.tsx` | 传递 displayOverride 到 startStream |
+| `src/lib/stream-session-manager.ts` | StartStreamParams 增加 displayOverride |
+
+### Phase 4: MCP 双文件管理
+
+| 文件 | 改动 |
+|------|------|
+| `src/app/api/plugins/mcp/route.ts` | POST 支持 SSE/HTTP；PUT 按 `_source` 分流写入；POST 检查双文件同名 |
+| `src/app/api/plugins/mcp/[name]/route.ts` | DELETE 检查双文件 |
+| `src/components/plugins/McpManager.tsx` | `MCPServerWithSource` 类型；保留 `_source` 标记 |
+| `src/components/plugins/McpServerEditor.tsx` | 远程服务器不生成空 command |
+
+### Phase 5: SDK Init 元数据转发
+
+| 文件 | 改动 |
+|------|------|
+| `src/lib/claude-client.ts` | system:init 转发 slash_commands/skills |
+| `src/hooks/useSSEStream.ts` | 新增 `onInitMeta` 回调 |
+| `src/lib/provider-resolver.ts` | `settingSources` 始终包含 `'user'` |
+
+### Phase 6: Completion 去重 + 去 localhost
+
+| 文件 | 改动 |
+|------|------|
+| `src/lib/onboarding-processor.ts` | **新增** — 从 onboarding route 提取核心逻辑 |
+| `src/lib/checkin-processor.ts` | **新增** — 从 checkin route 提取核心逻辑 |
+| `src/app/api/workspace/onboarding/route.ts` | 委托给 processOnboarding() |
+| `src/app/api/workspace/checkin/route.ts` | 委托给 processCheckin() |
+| `src/app/api/chat/route.ts` | processCompletionServerSide 直接调用处理函数，不走 HTTP |
+| `src/components/chat/ChatView.tsx` | 删除 detectAssistantCompletion — 仅保留服务端路径 |
+
+## 架构决策
+
+1. **Completion 单路径**：仅在服务端 `collectStreamResponse` 的 `finally` 块中处理，前端不再参与。理由：服务端路径在所有场景下都会执行（包括页面刷新、解析失败），前端路径是多余的且引入竞态。
+
+2. **处理函数直接导入**：`processOnboarding/processCheckin` 作为 lib 函数导出，route handler 和 completion detector 都直接调用，消除 HTTP 回调。
+
+3. **幂等保护**：`processOnboarding` 检查 `state.onboardingComplete`，`processCheckin` 检查 `state.lastCheckInDate === today`，即使意外多次调用也不会产生副作用。
+
+## 测试
+
+- 289 个单元测试全部通过
+- TypeScript 编译无错误
+- CDP 验证：Popover 三组分类正确，MCP 远程服务器创建/删除正常，Console 无 error