feat(vscode-ui): port /init and /compress slash commands from CLI

Until now vscode-ui-plugin only supported user-defined .toml slash commands; the 52 built-in commands shipped in CLI were absent from the visual UI. This commit establishes a small built-in command framework on top of slashCommandService and lights up the two highest-value commands first: /init and /compress.

Architecture:

  - SlashCommandInfo gains two optional fields: execution ('prompt' | 'side_effect') and sideEffect (currently 'compress'). TOML commands and built-in prompt commands continue to flow through processCommandPrompt and the existing 'webview turns prompt into user message' path, so all existing TOML behavior is preserved.

  - BUILT_IN_COMMANDS array is loaded BEFORE TOML directories, so a user's own ~/.deepv/commands/init.toml or compress.toml will override the built-in version (matches CLI command-resolution order).

  - getCommand() now also matches altNames (e.g. /summarize and /compact resolve to /compress), aligning with CLI compressCommand.

/init implementation:

  - INIT_COMMAND_PROMPT cloned from packages/cli/src/ui/commands/prompts/initPrompt.ts (header comment in initPrompt.ts marks the intentional duplication; if it diverges it should be lifted into core).

  - Backend handler creates an empty DEEPV.md when the file is missing or zero-bytes (mirrors CLI initCommand). When DEEPV.md already exists with content it returns an explicit error rather than silently overwriting the user's notes - a safer MVP than the CLI's init-choice dialog.

/compress implementation:

  - Side-effect command. The execute_custom_slash_command handler returns { sideEffect: 'compress', info } and MessageInput's slash dispatch detects this discriminator, suppresses the normal sendMessage path, and posts a builtin_compress message back to the extension.

  - The new builtin_compress handler resolves the current session's GeminiClient via sessionManager.getCurrentInitializedAIService() and calls geminiClient.tryCompressChat(promptId, signal, true) - byte-for-byte equivalent to packages/cli/src/ui/commands/compressCommand.ts.

  - Compression progress and result are surfaced via the existing show_notification channel; failures (compression already in progress, client not ready, etc.) are reported back to the user instead of being swallowed.

Tests: npm run build green across all 3 workspaces (core/cli/vscode-ui-plugin). Manual test plan: type /init in webview with no DEEPV.md → file created + INIT_COMMAND_PROMPT submitted to AI; type /compress on a long session → tryCompressChat invoked, result surfaced as a notification, no 'phantom AI message' sent.

Author: JerryliDe

packages/vscode-ui-plugin/src/constants/initPrompt.ts

@@ -0,0 +1,78 @@
+/**
+ * @license
+ * Copyright 2025 DeepV Code team
+ * https://github.com/OrionStarAI/DeepVCode
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+/**
+ * VSCode UI 端 /init 命令使用的提示词。
+ *
+ * ⚠️ 与 CLI 端 packages/cli/src/ui/commands/prompts/initPrompt.ts 的
+ *    INIT_COMMAND_PROMPT **保持文本一致**，避免两端行为漂移。如果以后修改，
+ *    两边都要改（或者后续再抽到 core 共享）。
+ *
+ * 该提示词由 /init 命令在 webview 端作为普通 user message 发送给 AI，
+ * AI 收到后会扫描整个工程并写入 DEEPV.md。
+ */
+
+export const INIT_COMMAND_PROMPT = `
+You are a DeepV Code AI assistant that brings the power of AI directly into the terminal. Your task is to deeply analyze the current directory and create a comprehensive DEEPV.md project context file.
+
+**Analysis Process:**
+
+1.  **Initial Exploration:**
+    *   Start by listing the files and directories to get a high-level overview of the structure.
+    *   Read the README file (e.g., \`README.md\`, \`README.txt\`) if it exists. This is often the best place to start.
+
+2.  **Launch Code Analysis Expert:**
+    *   Use the \`task\` tool to launch a specialized code analysis expert
+    *   Provide detailed instructions for comprehensive project analysis
+    *   Let the expert systematically explore the codebase, understand architecture, and analyze patterns
+    *   The expert should examine:
+        - Configuration files (\`package.json\`, \`tsconfig.json\`, \`webpack.config.js\`, etc.)
+        - Main source files, key modules, and entry points
+        - Documentation files, build scripts, and deployment configurations
+        - Directory structures and project architecture
+        - Test files to understand testing patterns and coverage
+        - Dependencies and external integrations
+
+3.  **Identify Project Type:**
+    *   **Code Project:** Look for clues like \`package.json\`, \`requirements.txt\`, \`pom.xml\`, \`go.mod\`, \`Cargo.toml\`, \`build.gradle\`, or a \`src\` directory. If you find them, this is likely a software project.
+    *   **Non-Code Project:** If you don't find code-related files, this might be a directory for documentation, research papers, notes, or something else.
+
+**Final Steps:**
+
+1. **Create DEEPV.md File:**
+   Based on the code analysis expert's findings and your own exploration, use the \`write_file\` tool to create a comprehensive DEEPV.md file in the current directory. The file should contain:
+
+   **For a Code Project:**
+   *   Project overview and purpose
+   *   Main technologies and framework architecture
+   *   Key build/run/test commands (from \`package.json\`, \`Makefile\`, etc.)
+   *   Development conventions and coding standards
+   *   Detailed file/directory structure explanation
+   *   Dependencies and external integrations
+   *   Special configurations or patterns
+   *   Testing strategies and CI/CD setup
+
+   **For a Non-Code Project:**
+   *   Directory purpose and contents overview
+   *   Key files and their purposes
+   *   Usage patterns and workflows
+   *   Organization structure
+
+2. **MANDATORY: Record Generation Info:**
+   After successfully creating the DEEPV.md file, you MUST use ONE \`save_memory\` call to record the generation timestamp. Use the current date and time:
+   \`\`\`
+   save_memory(fact="DEEPV.md generated by /init command on YYYY-MM-DD HH:mm:ss")
+   \`\`\`
+   Replace YYYY-MM-DD HH:mm:ss with the actual current date and time when you execute this.
+
+**Important Notes:**
+- This is for DeepV Code, a customized AI programming assistant
+- Be thorough in your analysis - explore the entire project comprehensively
+- Create a detailed, well-structured DEEPV.md file using write_file tool
+- Format the content as clear Markdown that will help future AI interactions
+- **CRITICAL**: You MUST use save_memory exactly once at the end to record the generation timestamp - this is mandatory and cannot be skipped
+`;

packages/vscode-ui-plugin/src/extension.ts

@@ -2958,7 +2958,94 @@ function setupSlashCommandHandlers() {
         return;
       }
 
-      // 处理命令的 prompt
+      // ─────────────────────────────────────────────────────────────
+      // 内置 side-effect 命令分发：不返回 prompt，而是执行后端动作
+      // ─────────────────────────────────────────────────────────────
+      if (command.execution === 'side_effect') {
+        if (command.sideEffect === 'compress') {
+          // /compress：让 webview 走副作用分支自己驱动压缩，与 CLI compressCommand
+          // 行为对齐（CLI 内部也是直接调 geminiClient.tryCompressChat）。
+          // 这里只把"我是 compress 副作用"反馈回去；真正调用 tryCompressChat 的
+          // 路径已经存在于 MultiSessionApp 里（compression_confirmation 流），
+          // webview 收到后转发一条 builtin_compress 消息回 backend 实际执行。
+          communicationService.sendMessage({
+            type: 'slash_command_result',
+            payload: {
+              success: true,
+              sideEffect: 'compress',
+              info: 'Compressing conversation history…',
+            },
+          });
+          return;
+        }
+        // 未来扩展其它 side-effect 类型时在此追加分支
+        communicationService.sendMessage({
+          type: 'slash_command_result',
+          payload: { success: false, error: `Unsupported side effect: ${command.sideEffect}` },
+        });
+        return;
+      }
+
+      // ─────────────────────────────────────────────────────────────
+      // /init 内置命令的特殊预处理：在 workspace 创建空 DEEPV.md，再走 prompt 分支
+      // ─────────────────────────────────────────────────────────────
+      if (command.kind === 'built-in' && command.name === 'init') {
+        const workspaceFolders = vscode.workspace.workspaceFolders;
+        if (!workspaceFolders || workspaceFolders.length === 0) {
+          communicationService.sendMessage({
+            type: 'slash_command_result',
+            payload: { success: false, error: '/init requires an open workspace folder.' },
+          });
+          return;
+        }
+        const workspaceRoot = workspaceFolders[0].uri.fsPath;
+        const deepvMdPath = path.join(workspaceRoot, 'DEEPV.md');
+        try {
+          let info: string | undefined;
+          if (fs.existsSync(deepvMdPath)) {
+            const stats = fs.statSync(deepvMdPath);
+            if (stats.size === 0) {
+              // 空文件：当不存在处理（与 CLI initCommand 行为对齐）
+              info = `Empty DEEPV.md detected at ${deepvMdPath}, regenerating…`;
+            } else {
+              // 非空文件：MVP 阶段保守拒绝执行，避免覆盖用户内容。
+              // CLI 端在这里会弹 init-choice 对话框；webview 端先走拒绝路径，
+              // 后续如有用户呼声再补对话框。
+              const sizeKB = Math.round(stats.size / 1024 * 100) / 100;
+              communicationService.sendMessage({
+                type: 'slash_command_result',
+                payload: {
+                  success: false,
+                  error: `DEEPV.md already exists (${sizeKB}KB). Delete or rename it first if you want to regenerate via /init.`,
+                },
+              });
+              return;
+            }
+          } else {
+            // 不存在：创建空文件
+            fs.writeFileSync(deepvMdPath, '', 'utf8');
+            info = `Created empty DEEPV.md at ${deepvMdPath}`;
+          }
+          const processedPrompt = slashCommandService.processCommandPrompt(command, args);
+          communicationService.sendMessage({
+            type: 'slash_command_result',
+            payload: { success: true, prompt: processedPrompt, info },
+          });
+          return;
+        } catch (initErr) {
+          logger.error('Failed to prepare DEEPV.md for /init', initErr instanceof Error ? initErr : undefined);
+          communicationService.sendMessage({
+            type: 'slash_command_result',
+            payload: {
+              success: false,
+              error: `Failed to prepare DEEPV.md: ${initErr instanceof Error ? initErr.message : String(initErr)}`,
+            },
+          });
+          return;
+        }
+      }
+
+      // 默认 prompt 模式（TOML 命令 / 其它 built-in prompt 命令）
       const processedPrompt = slashCommandService.processCommandPrompt(command, args);
 
       communicationService.sendMessage({
@@ -2974,6 +3061,70 @@ function setupSlashCommandHandlers() {
     }
   });
 
+  // ─────────────────────────────────────────────────────────────────
+  // 内置 /compress 的实际执行入口：webview 收到 sideEffect:'compress' 后
+  // 会回发一条 builtin_compress 消息触发这里。我们直接调用 core 的
+  // tryCompressChat（与 CLI compressCommand 完全一致），并把结果
+  // 通过现有 system-notification / chat_compressed 通道反馈给 webview。
+  // ─────────────────────────────────────────────────────────────────
+  communicationService.addMessageHandler('builtin_compress', async () => {
+    try {
+      // 当前会话的 AIService —— 与 setupChatHandlers 内 send_message 的
+      // 取法（getCurrentInitializedAIService）保持一致。
+      const aiService = await sessionManager.getCurrentInitializedAIService();
+      const geminiClient = aiService?.getGeminiClient?.();
+      if (!geminiClient) {
+        logger.warn('[/compress] geminiClient not initialized');
+        communicationService.sendMessage({
+          type: 'slash_command_result',
+          payload: { success: false, error: 'AI client not ready. Please wait for initialization.' },
+        });
+        return;
+      }
+
+      if (geminiClient.isCompressionInProgress?.()) {
+        communicationService.sendMessage({
+          type: 'slash_command_result',
+          payload: { success: false, error: 'Compression already in progress, please wait.' },
+        });
+        return;
+      }
+
+      logger.info('[/compress] Manual compression triggered via slash command');
+      const promptId = `slash-compress-${Date.now()}`;
+      const result = await geminiClient.tryCompressChat(promptId, new AbortController().signal, true);
+
+      if (result) {
+        const before = (result as any).originalTokenCount;
+        const after = (result as any).newTokenCount;
+        logger.info(`[/compress] Compressed ${before} → ${after} tokens`);
+        const fmt = (n: any) => (typeof n === 'number' ? n.toLocaleString() : String(n));
+        communicationService.sendMessage({
+          type: 'slash_command_result',
+          payload: {
+            success: true,
+            info: `Compressed ${fmt(before)} → ${fmt(after)} tokens`,
+          },
+        });
+      } else {
+        logger.warn('[/compress] tryCompressChat returned falsy result');
+        communicationService.sendMessage({
+          type: 'slash_command_result',
+          payload: { success: false, error: 'Compression returned no result. The history may already be small enough.' },
+        });
+      }
+    } catch (error) {
+      logger.error('[/compress] Compression failed', error instanceof Error ? error : undefined);
+      communicationService.sendMessage({
+        type: 'slash_command_result',
+        payload: {
+          success: false,
+          error: `Compression failed: ${error instanceof Error ? error.message : String(error)}`,
+        },
+      });
+    }
+  });
+
   logger.info('🎯 Slash command handlers registered');
 }
 

packages/vscode-ui-plugin/src/services/slashCommandService.ts

@@ -10,7 +10,16 @@
  * Loads custom slash commands from .toml files, sharing the same configuration
  * paths as the CLI (~/.deepv/commands and <project>/.deepvcode/commands).
  *
- * This is a simplified version of CLI's FileCommandLoader, adapted for VSCode environment.
+ * Also exposes a small set of **built-in** commands that mirror high-value
+ * CLI counterparts (currently: /init, /compress).
+ *
+ * Built-in commands have two execution shapes:
+ *   1. `prompt` — same as TOML commands: returns a processed prompt string,
+ *     the webview then sends it to the AI as a normal user message.
+ *   2. `side_effect` — the command performs a backend action (e.g. compress
+ *     history) and the webview MUST NOT forward it to the AI. Side-effect
+ *     commands carry a discriminator (`sideEffect: 'compress' | …`) which
+ *     the webview switches on to dispatch the right local action.
  */
 
 import * as fs from 'fs/promises';
@@ -20,6 +29,7 @@ import * as toml from '@iarna/toml';
 import { glob } from 'glob';
 import { getUserCommandsDirs, getProjectCommandsDirs } from 'deepv-code-core';
 import { Logger } from '../utils/logger';
+import { INIT_COMMAND_PROMPT } from '../constants/initPrompt';
 
 /**
  * Slash command info sent to webview (serializable)
@@ -31,8 +41,20 @@ export interface SlashCommandInfo {
   description: string;
   /** Command source: 'file' for custom commands, 'built-in' for hardcoded */
   kind: 'file' | 'built-in';
-  /** The prompt template */
+  /** The prompt template (TOML 命令直接是用户写的；built-in 'prompt' 命令是固定文本；side-effect 命令为空字符串) */
   prompt: string;
+  /**
+   * Built-in 命令的执行风格：
+   *   - undefined / 'prompt'：和 TOML 命令一样，processCommandPrompt 返回字符串，
+   *     webview 拿去当 user message 发给 AI。
+   *   - 'side_effect'：调用方应当走副作用分支（不发 AI），由 sideEffect 字段
+   *     指明执行哪种动作。
+   */
+  execution?: 'prompt' | 'side_effect';
+  /** 副作用类型（execution === 'side_effect' 时必填） */
+  sideEffect?: 'compress';
+  /** 命令别名（webview 在 fuzzy 匹配 / 输入解析时会兼容这些别名） */
+  altNames?: string[];
 }
 
 /**
@@ -48,6 +70,36 @@ interface TomlCommandDef {
  */
 const SHORTHAND_ARGS_PLACEHOLDER = '{{args}}';
 
+/**
+ * 内置命令清单。**故意**和 CLI `packages/cli/src/ui/commands/*.ts` 保持
+ * 字段语义一致：name / altNames / description 都从 CLI 复制过来，让用户
+ * 在两端体验一致。
+ *
+ * 现仅注册 /init 与 /compress 两个最高频的命令；后续如需移植 /clear /memory
+ * 等命令，只需在这里追加新条目并在 setupSlashCommandHandlers 中扩展副作用
+ * 分支即可。
+ */
+const BUILT_IN_COMMANDS: SlashCommandInfo[] = [
+  {
+    name: 'init',
+    description: 'Analyze the current workspace and create a DEEPV.md project context file',
+    kind: 'built-in',
+    execution: 'prompt',
+    prompt: INIT_COMMAND_PROMPT,
+  },
+  {
+    name: 'compress',
+    altNames: ['summarize', 'compact'],
+    description: 'Manually compress the conversation history to free up context (calls tryCompressChat)',
+    kind: 'built-in',
+    execution: 'side_effect',
+    sideEffect: 'compress',
+    // side-effect 命令不会被作为 prompt 发送，但保留一段说明文本，
+    // 以便万一调用方误用 processCommandPrompt 时仍有合理 fallback。
+    prompt: '[Built-in /compress invoked]',
+  },
+];
+
 /**
  * Service for managing custom slash commands in VSCode
  */
@@ -91,10 +143,20 @@ export class SlashCommandService {
   }
 
   /**
-   * Get a specific command by name
+   * Get a specific command by name. 同时匹配主名 + altNames（CLI 行为对齐）。
    */
   getCommand(name: string): SlashCommandInfo | undefined {
-    return this.commands.get(name);
+    if (!name) return undefined;
+    // 直接命中主名（最常见路径）
+    const direct = this.commands.get(name);
+    if (direct) return direct;
+    // 退化：扫一次 altNames 命中（命令数量很少，O(n) 完全可接受）
+    for (const cmd of this.commands.values()) {
+      if (cmd.altNames && cmd.altNames.includes(name)) {
+        return cmd;
+      }
+    }
+    return undefined;
   }
 
   /**
@@ -119,8 +181,21 @@ export class SlashCommandService {
 
   /**
    * Load commands from both user and project directories
+   *
+   * 加载顺序（后者覆盖前者）：
+   *   1) Built-in 命令（/init, /compress …）
+   *   2) 用户级 TOML 命令
+   *   3) 项目级 TOML 命令
+   *
+   * 用户的 TOML 命令可以覆盖 built-in（比如想给 /compress 自定义行为），
+   * 这与 CLI 的命令解析优先级保持一致。
    */
   private async loadCommands(): Promise<void> {
+    // 1) 先注册 built-in 命令
+    for (const cmd of BUILT_IN_COMMANDS) {
+      this.commands.set(cmd.name, cmd);
+    }
+
     const globOptions = {
       nodir: true,
       dot: true,

packages/vscode-ui-plugin/src/types/messages.ts

@@ -500,7 +500,16 @@ export type ExtensionToWebViewMessage =
   | { type: 'stream_recovery_end'; payload: { sessionId: string } }
   // 🎯 自定义斜杠命令相关
   | { type: 'slash_commands_list'; payload: { commands: SlashCommandInfo[] } }
-  | { type: 'slash_command_result'; payload: { success: boolean; prompt?: string; error?: string } }
+  | { type: 'slash_command_result'; payload: {
+      success: boolean;
+      /** Prompt 模式：返回处理后的 prompt 给 webview 转发为 user message */
+      prompt?: string;
+      /** Side-effect 模式：webview 检测到此字段则不发 AI，转发对应消息给 backend 触发动作 */
+      sideEffect?: 'compress';
+      /** 信息提示：用于在 webview 上显示一条系统通知（成功 / 跳过等） */
+      info?: string;
+      error?: string;
+    } }
   // 🎯 模型切换压缩确认
   | { type: 'compression_confirmation_request'; payload: { requestId: string; sessionId: string; targetModel: string; currentTokens: number; targetTokenLimit: number; compressionThreshold: number; message: string } }
   // 🎯 Token使用情况更新（压缩后）