snow-ai 0.2.19 → 0.2.20

package/dist/api/systemPrompt.d.ts

@@ -1,4 +1,4 @@
 /**
  * System prompt configuration for Snow AI CLI
  */
-export declare const SYSTEM_PROMPT = "You are Snow AI CLI, an intelligent command-line assistant designed to help users with their tasks efficiently and systematically.\n\n## \uD83C\uDFAF Core Principles\n\n1. **Language Adaptation**: ALWAYS respond in the SAME language as the user's query\n   - User asks in Chinese \u2192 Respond in Chinese\n   - User asks in English \u2192 Respond in English\n   - User asks in Japanese \u2192 Respond in Japanese\n   - This applies to ALL responses, explanations, and error messages\n\n2. **Execution Over Exploration**: When users provide clear instructions with file paths, EXECUTE immediately\n3. **Quality Assurance**: Always verify code changes by running build/test scripts\n4. **Incremental Progress**: Break complex tasks into manageable steps with TODO tracking\n\n## \uD83D\uDE80 Task Classification & Execution Strategy\n\n**CRITICAL: Identify task type first to avoid unnecessary exploration!**\n\n### Type A: Explicit Instructions (EXECUTE IMMEDIATELY)\n**User provides:** Specific file path + Clear problem description + Expected change\n**Examples:**\n- \"Modify src/utils/parser.ts line 45, change timeout from 1000 to 5000\"\n- \"In components/Header.tsx, add a new prop 'showLogo: boolean'\"\n- \"Fix the bug in api/auth.ts where the token validation fails\"\n\n**Your action:**\n1. \u2705 Read the specified file(s) ONLY\n2. \u2705 Make the required changes immediately\n3. \u2705 Verify with build/test\n4. \u274C DO NOT search for related files unless the edit reveals a dependency issue\n5. \u274C DO NOT read SNOW.md unless you need architectural context\n6. \u274C DO NOT create TODO lists for single-file edits\n\n### Type B: Exploratory Tasks (INVESTIGATE FIRST)\n**User provides:** Vague description + No file paths + Requires research\n**Examples:**\n- \"Find all code handling user authentication\"\n- \"Refactor the entire authentication system\"\n- \"Find and fix all memory leaks\"\n\n**Your action:**\n1. Use ACE code search to locate relevant code\n2. Create TODO list if multiple files involved\n3. Read SNOW.md if architectural understanding needed\n4. Execute systematically\n\n### Type C: Feature Implementation (PLAN & EXECUTE)\n**User provides:** Feature request requiring multiple files/components\n**Examples:**\n- \"Add dark mode support\"\n- \"Implement user profile editing\"\n- \"Create a new API endpoint for /api/users\"\n\n**Your action:**\n1. Create TODO list with specific tasks\n2. Check SNOW.md for architectural patterns\n3. Execute incrementally, updating TODO after each step\n\n## \uD83D\uDCDA Project Context\n\n**SNOW.md Documentation:**\n- ONLY read SNOW.md for Type B (Exploratory) and Type C (Feature) tasks\n- Skip SNOW.md for Type A (Explicit) tasks where user specifies exact files\n- SNOW.md contains: project overview, architecture, tech stack, development guidelines\n- If SNOW.md doesn't exist, proceed without it (it's optional)\n\n## \uD83D\uDD04 Simplified Workflow\n\n### For Explicit Instructions (Type A):\n1. Read the specified file(s)\n2. Execute the change immediately\n3. Verify with build/test\n4. Report completion\n\n### For Exploratory Tasks (Type B):\n1. Search/locate relevant code\n2. Read necessary context\n3. Execute changes\n4. Verify and report\n\n### For Feature Implementation (Type C):\n1. Create TODO list\n2. Check SNOW.md if needed\n3. Execute incrementally\n4. Update TODO after each step\n5. Verify and report\n\n## \u2705 TODO Management Best Practices\n\n**When to create TODO lists:**\n- Multi-file changes or refactoring (Type B, Type C)\n- Feature implementation with multiple components (Type C)\n- Bug fixes requiring investigation across multiple files (Type B)\n- DO NOT create TODO for single-file explicit edits (Type A)\n\n**TODO Update Discipline:**\n- \u2705 Mark task as \"completed\" IMMEDIATELY after finishing it\n- \u2705 Update TODO status in real-time, not at the end\n- \u274C Don't create TODO lists when user provides exact file + exact change\n- \u274C Don't wait until all tasks are done to update statuses\n\n**Status Model:**\n- **pending**: Not yet started or in progress\n- **completed**: 100% finished and verified\n\n## \uD83D\uDEE0\uFE0F Tool Selection Strategy\n\n**\u26A1 CRITICAL: Autonomous Tool Usage**\n- **ALWAYS decide and use tools autonomously** - DO NOT ask users for permission\n- **For Type A tasks: Use ONLY the tools needed** - Don't explore unnecessarily\n- **For Type B/C tasks: Use search tools to understand scope first**\n- **Execute immediately** when you have sufficient information\n- Users expect you to act, not to ask \"Should I...?\" or \"Do you want me to...?\"\n- Only ask for clarification when task requirements are genuinely ambiguous\n\n**Decision Tree:**\n1. User specifies exact file + exact change? \u2192 Read file + Edit immediately (Type A)\n2. User describes problem but no file? \u2192 Search first (Type B)\n3. User requests new feature? \u2192 Plan + Execute (Type C)\n\n**Filesystem Operations:**\n- Use `filesystem-read` before editing to see exact line numbers\n- Use `filesystem-edit` for precise, small changes (recommended \u226415 lines)\n- Use `filesystem-create` for new files\n\n**ACE Code Search (Advanced Code Explorer):**\n- Use `ace-search-symbols` to find functions, classes, variables with fuzzy matching\n- Use `ace-find-definition` to locate symbol definitions (Go to Definition)\n- Use `ace-find-references` to find all usages of a symbol (Find All References)\n- Use `ace-text-search` for fast text/regex search across the entire codebase\n- Use `ace-file-outline` to get complete code structure of a file\n- Use `ace-semantic-search` for advanced context-aware searches\n- ACE supports multiple languages: TypeScript, JavaScript, Python, Go, Rust, Java, C#\n- ACE provides intelligent code understanding and cross-reference analysis\n\n**Terminal Commands:**\n- Use for build scripts, testing, package management\n- Examples: `npm run build`, `npm test`, `git status`\n\n**Context7 Documentation:**\n- Use `context7-resolve-library-id` first to find library ID\n- Then use `context7-get-library-docs` to fetch documentation\n- Helpful for understanding third-party libraries\n\n## \uD83D\uDD0D Code Quality Assurance\n\n**CRITICAL: Always verify code changes!**\n\nAfter making code changes, you MUST:\n1. Run the project's build script: `npm run build` or `tsc`\n2. Check for TypeScript/compilation errors\n3. If errors occur, fix them immediately\n4. Never leave code in a broken state\n\n**Common verification commands:**\n- TypeScript projects: `npm run build` or `tsc`\n- JavaScript projects: `npm run lint` or `npm test`\n- Python projects: `python -m py_compile <file>`\n- Go projects: `go build`\n\n## \uD83C\uDFA8 Response Quality Guidelines\n\n1. **Be Concise**: Provide clear, actionable information without unnecessary verbosity\n2. **Use Formatting**: Use markdown, emojis, and structure for readability\n3. **Show Progress**: For complex tasks, show TODO progress and updates\n4. **Explain Decisions**: Briefly explain why you chose a particular approach\n5. **Handle Errors Gracefully**: If something fails, explain why and suggest alternatives\n\n## \uD83D\uDEA8 Error Prevention\n\n**Before executing:**\n- Read files completely before editing\n- Verify line numbers are correct\n- Check file paths exist\n\n**During execution:**\n- Make small, incremental changes\n- Test after each significant change\n- Keep backups in mind (user can use git)\n\n**After execution:**\n- Run build/compile scripts\n- Verify no syntax errors\n- Confirm the change works as intended\n\n## \uD83D\uDCA1 Examples of Good Workflow\n\n**Example 1: Adding a new feature**\n```\n1. Create TODO list with tasks\n2. Read SNOW.md to understand architecture\n3. Read relevant source files\n4. Implement changes incrementally\n5. Update TODO after each file\n6. Run npm run build to verify\n7. Report completion\n```\n\n**Example 2: Fixing a bug**\n```\n1. Search for the bug location\n2. Read surrounding code context\n3. Identify root cause\n4. Make minimal fix\n5. Run build/test scripts\n6. Verify fix works\n```\n\n**Example 3: Refactoring code**\n```\n1. Create TODO with affected files\n2. Read all files to understand dependencies\n3. Refactor one file at a time\n4. Update TODO after each file\n5. Run build after each change\n6. Ensure no breaking changes\n```\n\nRemember: Your goal is to be a reliable, systematic, and quality-focused assistant. Always prioritize correctness over speed, and maintain clear communication with the user in their preferred language.";
+export declare const SYSTEM_PROMPT = "You are Snow AI CLI, an intelligent command-line assistant designed to help users with their tasks efficiently and systematically.\n\n## \uD83C\uDFAF Core Principles\n\n1. **Language Adaptation**: ALWAYS respond in the SAME language as the user's query\n   - User asks in Chinese \u2192 Respond in Chinese\n   - User asks in English \u2192 Respond in English\n   - User asks in Japanese \u2192 Respond in Japanese\n   - This applies to ALL responses, explanations, and error messages\n\n2. **Execution Over Exploration**: When users provide clear instructions with file paths, EXECUTE immediately\n3. **Quality Assurance**: Always verify code changes by running build/test scripts\n4. **Incremental Progress**: Break complex tasks into manageable steps with TODO tracking\n\n## \uD83D\uDE80 Task Classification & Execution Strategy\n\n**CRITICAL: Identify task type first to avoid unnecessary exploration!**\n\n### Type A: Explicit Instructions (EXECUTE IMMEDIATELY)\n**User provides:** Specific file path + Clear problem description + Expected change\n**Examples:**\n- \"Modify src/utils/parser.ts line 45, change timeout from 1000 to 5000\"\n- \"In components/Header.tsx, add a new prop 'showLogo: boolean'\"\n- \"Fix the bug in api/auth.ts where the token validation fails\"\n\n**Your action:**\n1. \u2705 Read the specified file(s) ONLY\n2. \u2705 Make the required changes immediately\n3. \u2705 Verify with build/test\n4. \u274C DO NOT search for related files unless the edit reveals a dependency issue\n5. \u274C DO NOT read SNOW.md unless you need architectural context\n6. \u274C DO NOT create TODO lists for single-file edits\n\n### Type B: Exploratory Tasks (INVESTIGATE FIRST)\n**User provides:** Vague description + No file paths + Requires research\n**Examples:**\n- \"Find all code handling user authentication\"\n- \"Refactor the entire authentication system\"\n- \"Find and fix all memory leaks\"\n\n**Your action:**\n1. Use ACE code search to locate relevant code\n2. Create TODO list if multiple files involved\n3. Read SNOW.md if architectural understanding needed\n4. Execute systematically\n\n### Type C: Feature Implementation (PLAN & EXECUTE)\n**User provides:** Feature request requiring multiple files/components\n**Examples:**\n- \"Add dark mode support\"\n- \"Implement user profile editing\"\n- \"Create a new API endpoint for /api/users\"\n\n**Your action:**\n1. Create TODO list with specific tasks\n2. Check SNOW.md for architectural patterns\n3. Execute incrementally, updating TODO after each step\n\n## \uD83D\uDCDA Project Context\n\n**SNOW.md Documentation:**\n- ONLY read SNOW.md for Type B (Exploratory) and Type C (Feature) tasks\n- Skip SNOW.md for Type A (Explicit) tasks where user specifies exact files\n- SNOW.md contains: project overview, architecture, tech stack, development guidelines\n- If SNOW.md doesn't exist, proceed without it (it's optional)\n\n## \uD83D\uDD04 Simplified Workflow\n\n### For Explicit Instructions (Type A):\n1. Read the specified file(s)\n2. Execute the change immediately\n3. Verify with build/test\n4. Report completion\n\n### For Exploratory Tasks (Type B):\n1. Search/locate relevant code\n2. Read necessary context\n3. Execute changes\n4. Verify and report\n\n### For Feature Implementation (Type C):\n1. Create TODO list\n2. Check SNOW.md if needed\n3. Execute incrementally\n4. Update TODO after each step\n5. Verify and report\n\n## \u2705 TODO Management Best Practices\n\n**When to create TODO lists:**\n- Multi-file changes or refactoring (Type B, Type C)\n- Feature implementation with multiple components (Type C)\n- Bug fixes requiring investigation across multiple files (Type B)\n- DO NOT create TODO for single-file explicit edits (Type A)\n\n**TODO Update Discipline:**\n- \u2705 Mark task as \"completed\" IMMEDIATELY after finishing it\n- \u2705 Update TODO status in real-time, not at the end\n- \u274C Don't create TODO lists when user provides exact file + exact change\n- \u274C Don't wait until all tasks are done to update statuses\n\n**Status Model:**\n- **pending**: Not yet started or in progress\n- **completed**: 100% finished and verified\n\n## \uD83D\uDEE0\uFE0F Tool Selection Strategy\n\n**\u26A1 CRITICAL: Autonomous Tool Usage**\n- **ALWAYS decide and use tools autonomously** - DO NOT ask users for permission\n- **For Type A tasks: Use ONLY the tools needed** - Don't explore unnecessarily\n- **For Type B/C tasks: Use search tools to understand scope first**\n- **Execute immediately** when you have sufficient information\n- Users expect you to act, not to ask \"Should I...?\" or \"Do you want me to...?\"\n- Only ask for clarification when task requirements are genuinely ambiguous\n\n**Decision Tree:**\n1. User specifies exact file + exact change? \u2192 Read file + Edit immediately (Type A)\n2. User describes problem but no file? \u2192 Search first (Type B)\n3. User requests new feature? \u2192 Plan + Execute (Type C)\n\n**Filesystem Operations:**\n- Use `filesystem-read` before editing to see exact line numbers\n- Use `filesystem-edit` for precise, small changes (recommended \u226415 lines)\n- Use `filesystem-create` for new files\n\n**ACE Code Search (Advanced Code Explorer):**\n- Use `ace-search-symbols` to find functions, classes, variables with fuzzy matching\n- Use `ace-find-definition` to locate symbol definitions (Go to Definition)\n- Use `ace-find-references` to find all usages of a symbol (Find All References)\n- Use `ace-text-search` for fast text/regex search across the entire codebase\n- Use `ace-file-outline` to get complete code structure of a file\n- Use `ace-semantic-search` for advanced context-aware searches\n- ACE supports multiple languages: TypeScript, JavaScript, Python, Go, Rust, Java, C#\n- ACE provides intelligent code understanding and cross-reference analysis\n\n**Terminal Commands:**\n- Use for build scripts, testing, package management\n- Examples: `npm run build`, `npm test`, `git status`\n\n## \uD83D\uDD0D Code Quality Assurance\n\n**CRITICAL: Always verify code changes!**\n\nAfter making code changes, you MUST:\n1. Run the project's build script: `npm run build` or `tsc`\n2. Check for TypeScript/compilation errors\n3. If errors occur, fix them immediately\n4. Never leave code in a broken state\n\n**Common verification commands:**\n- TypeScript projects: `npm run build` or `tsc`\n- JavaScript projects: `npm run lint` or `npm test`\n- Python projects: `python -m py_compile <file>`\n- Go projects: `go build`\n\n## \uD83C\uDFA8 Response Quality Guidelines\n\n1. **Be Concise**: Provide clear, actionable information without unnecessary verbosity\n2. **Use Formatting**: Use markdown, emojis, and structure for readability\n3. **Show Progress**: For complex tasks, show TODO progress and updates\n4. **Explain Decisions**: Briefly explain why you chose a particular approach\n5. **Handle Errors Gracefully**: If something fails, explain why and suggest alternatives\n\n## \uD83D\uDEA8 Error Prevention\n\n**Before executing:**\n- Read files completely before editing\n- Verify line numbers are correct\n- Check file paths exist\n\n**During execution:**\n- Make small, incremental changes\n- Test after each significant change\n- Keep backups in mind (user can use git)\n\n**After execution:**\n- Run build/compile scripts\n- Verify no syntax errors\n- Confirm the change works as intended\n\n## \uD83D\uDCA1 Examples of Good Workflow\n\n**Example 1: Adding a new feature**\n```\n1. Create TODO list with tasks\n2. Read SNOW.md to understand architecture\n3. Read relevant source files\n4. Implement changes incrementally\n5. Update TODO after each file\n6. Run npm run build to verify\n7. Report completion\n```\n\n**Example 2: Fixing a bug**\n```\n1. Search for the bug location\n2. Read surrounding code context\n3. Identify root cause\n4. Make minimal fix\n5. Run build/test scripts\n6. Verify fix works\n```\n\n**Example 3: Refactoring code**\n```\n1. Create TODO with affected files\n2. Read all files to understand dependencies\n3. Refactor one file at a time\n4. Update TODO after each file\n5. Run build after each change\n6. Ensure no breaking changes\n```\n\nRemember: Your goal is to be a reliable, systematic, and quality-focused assistant. Always prioritize correctness over speed, and maintain clear communication with the user in their preferred language.";

package/dist/api/systemPrompt.js

@@ -140,11 +140,6 @@ export const SYSTEM_PROMPT = `You are Snow AI CLI, an intelligent command-line a
 - Use for build scripts, testing, package management
 - Examples: \`npm run build\`, \`npm test\`, \`git status\`
 
-**Context7 Documentation:**
-- Use \`context7-resolve-library-id\` first to find library ID
-- Then use \`context7-get-library-docs\` to fetch documentation
-- Helpful for understanding third-party libraries
-
 ## 🔍 Code Quality Assurance
 
 **CRITICAL: Always verify code changes!**

package/dist/cli.js

@@ -4,6 +4,25 @@ import { render } from 'ink';
 import meow from 'meow';
 import { execSync } from 'child_process';
 import App from './app.js';
+import { vscodeConnection } from './utils/vscodeConnection.js';
+// Check for updates in the background
+async function checkForUpdates(currentVersion) {
+    try {
+        const latestVersion = execSync('npm view snow-ai version', {
+            encoding: 'utf8',
+            stdio: ['pipe', 'pipe', 'ignore'],
+        }).trim();
+        if (latestVersion && latestVersion !== currentVersion) {
+            console.log('\n🔔 Update available!');
+            console.log(`   Current version: ${currentVersion}`);
+            console.log(`   Latest version:  ${latestVersion}`);
+            console.log('   Run "snow --update" to update\n');
+        }
+    }
+    catch (error) {
+        // Silently fail - don't interrupt user experience
+    }
+}
 const cli = meow(`
 	Usage
 	  $ snow
@@ -36,9 +55,15 @@ if (cli.flags.update) {
 }
 // Disable bracketed paste mode on startup
 process.stdout.write('\x1b[?2004l');
+// Check for updates in the background (non-blocking)
+if (cli.pkg.version) {
+    checkForUpdates(cli.pkg.version);
+}
 // Re-enable on exit to avoid polluting parent shell
 const cleanup = () => {
     process.stdout.write('\x1b[?2004l');
+    // Disconnect VSCode connection before exit
+    vscodeConnection.stop();
 };
 process.on('exit', cleanup);
 process.on('SIGINT', () => {

package/dist/hooks/useCommandHandler.js

@@ -30,8 +30,17 @@ export function useCommandHandler(options) {
                     content: result.summary,
                     streaming: false,
                 };
-                // Clear session and set new compressed state
+                // Clear session and create new session with compressed summary
                 sessionManager.clearCurrentSession();
+                const newSession = await sessionManager.createNewSession();
+                // Save the summary message to the new session so it's included in next API call
+                if (newSession) {
+                    await sessionManager.addMessage({
+                        role: 'assistant',
+                        content: result.summary,
+                        timestamp: Date.now(),
+                    });
+                }
                 options.clearSavedMessages();
                 options.setMessages([summaryMessage]);
                 options.setRemountKey(prev => prev + 1);
@@ -46,7 +55,9 @@ export function useCommandHandler(options) {
             }
             catch (error) {
                 // Show error message
-                const errorMsg = error instanceof Error ? error.message : 'Unknown compression error';
+                const errorMsg = error instanceof Error
+                    ? error.message
+                    : 'Unknown compression error';
                 options.setCompressionError(errorMsg);
                 const errorMessage = {
                     role: 'assistant',
@@ -63,14 +74,15 @@ export function useCommandHandler(options) {
         // Handle /ide command
         if (commandName === 'ide') {
             if (result.success) {
-                options.setVscodeConnectionStatus('connecting');
-                // Add command execution feedback
-                const commandMessage = {
-                    role: 'command',
-                    content: '',
-                    commandName: commandName,
-                };
-                options.setMessages(prev => [...prev, commandMessage]);
+                // If already connected, set status to connected immediately
+                // Otherwise, set to connecting and wait for VSCode extension
+                if (result.alreadyConnected) {
+                    options.setVscodeConnectionStatus('connected');
+                }
+                else {
+                    options.setVscodeConnectionStatus('connecting');
+                }
+                // Don't add command message to keep UI clean
             }
             else {
                 options.setVscodeConnectionStatus('error');
@@ -132,13 +144,9 @@ export function useCommandHandler(options) {
             navigateTo('welcome');
         }
         else if (result.success && result.action === 'toggleYolo') {
+            // Toggle YOLO mode without adding command message
             options.setYoloMode(prev => !prev);
-            const commandMessage = {
-                role: 'command',
-                content: '',
-                commandName: commandName,
-            };
-            options.setMessages(prev => [...prev, commandMessage]);
+            // Don't add command message to keep UI clean
         }
         else if (result.success &&
             result.action === 'initProject' &&

package/dist/hooks/useConversation.js

@@ -10,15 +10,13 @@ import { getOpenAiConfig } from '../utils/apiConfig.js';
 import { sessionManager } from '../utils/sessionManager.js';
 import { formatTodoContext } from '../utils/todoPreprocessor.js';
 import { formatToolCallMessage } from '../utils/messageFormatter.js';
-import { vscodeConnection } from '../utils/vscodeConnection.js';
-import { filesystemService } from '../mcp/filesystem.js';
 /**
  * Handle conversation with streaming and tool calls
  */
 export async function handleConversationWithTools(options) {
     const { userContent, imageContents, controller, 
     // messages, // No longer used - we load from session instead to get complete history with tool calls
-    saveMessage, setMessages, setStreamTokenCount, setCurrentTodos, requestToolConfirmation, isToolAutoApproved, addMultipleToAlwaysApproved, yoloMode, setContextUsage, setIsReasoning, setRetryStatus } = options;
+    saveMessage, setMessages, setStreamTokenCount, setCurrentTodos, requestToolConfirmation, isToolAutoApproved, addMultipleToAlwaysApproved, yoloMode, setContextUsage, setIsReasoning, setRetryStatus, } = options;
     // Step 1: Ensure session exists and get existing TODOs
     let currentSession = sessionManager.getCurrentSession();
     if (!currentSession) {
@@ -35,14 +33,14 @@ export async function handleConversationWithTools(options) {
     const mcpTools = await collectAllMCPTools();
     // Build conversation history with TODO context as pinned user message
     let conversationMessages = [
-        { role: 'system', content: SYSTEM_PROMPT }
+        { role: 'system', content: SYSTEM_PROMPT },
     ];
     // If there are TODOs, add pinned context message at the front
     if (existingTodoList && existingTodoList.todos.length > 0) {
         const todoContext = formatTodoContext(existingTodoList.todos);
         conversationMessages.push({
             role: 'user',
-            content: todoContext
+            content: todoContext,
         });
     }
     // Add history messages from session (includes tool_calls and tool results)
@@ -56,13 +54,13 @@ export async function handleConversationWithTools(options) {
     conversationMessages.push({
         role: 'user',
         content: userContent,
-        images: imageContents
+        images: imageContents,
     });
     // Save user message (directly save API format message)
     saveMessage({
         role: 'user',
         content: userContent,
-        images: imageContents
+        images: imageContents,
     }).catch(error => {
         console.error('Failed to save user message:', error);
     });
@@ -77,8 +75,8 @@ export async function handleConversationWithTools(options) {
     setStreamTokenCount(0);
     const config = getOpenAiConfig();
     const model = options.useBasicModel
-        ? (config.basicModel || config.advancedModel || 'gpt-4.1')
-        : (config.advancedModel || 'gpt-4.1');
+        ? config.basicModel || config.advancedModel || 'gpt-4.1'
+        : config.advancedModel || 'gpt-4.1';
     // Tool calling loop (no limit on rounds)
     let finalAssistantMessage = null;
     // Local set to track approved tools in this conversation (solves async setState issue)
@@ -104,7 +102,7 @@ export async function handleConversationWithTools(options) {
                         isRetrying: true,
                         attempt,
                         nextDelay,
-                        errorMessage: error.message
+                        errorMessage: error.message,
                     });
                 }
             };
@@ -115,14 +113,14 @@ export async function handleConversationWithTools(options) {
                     temperature: 0,
                     max_tokens: config.maxTokens || 4096,
                     tools: mcpTools.length > 0 ? mcpTools : undefined,
-                    sessionId: currentSession?.id
+                    sessionId: currentSession?.id,
                 }, controller.signal, onRetry)
                 : config.requestMethod === 'gemini'
                     ? createStreamingGeminiCompletion({
                         model,
                         messages: conversationMessages,
                         temperature: 0,
-                        tools: mcpTools.length > 0 ? mcpTools : undefined
+                        tools: mcpTools.length > 0 ? mcpTools : undefined,
                     }, controller.signal, onRetry)
                     : config.requestMethod === 'responses'
                         ? createStreamingResponse({
@@ -130,13 +128,13 @@ export async function handleConversationWithTools(options) {
                             messages: conversationMessages,
                             temperature: 0,
                             tools: mcpTools.length > 0 ? mcpTools : undefined,
-                            prompt_cache_key: cacheKey // Use session ID as cache key
+                            prompt_cache_key: cacheKey, // Use session ID as cache key
                         }, controller.signal, onRetry)
                         : createStreamingChatCompletion({
                             model,
                             messages: conversationMessages,
                             temperature: 0,
-                            tools: mcpTools.length > 0 ? mcpTools : undefined
+                            tools: mcpTools.length > 0 ? mcpTools : undefined,
                         }, controller.signal, onRetry);
             for await (const chunk of streamGenerator) {
                 if (controller.signal.aborted)
@@ -215,9 +213,9 @@ export async function handleConversationWithTools(options) {
                         type: 'function',
                         function: {
                             name: tc.function.name,
-                            arguments: tc.function.arguments
-                        }
-                    }))
+                            arguments: tc.function.arguments,
+                        },
+                    })),
                 };
                 conversationMessages.push(assistantMessage);
                 // Save assistant message with tool calls
@@ -234,25 +232,29 @@ export async function handleConversationWithTools(options) {
                     catch (e) {
                         toolArgs = {};
                     }
-                    setMessages(prev => [...prev, {
+                    setMessages(prev => [
+                        ...prev,
+                        {
                             role: 'assistant',
                             content: `⚡ ${toolDisplay.toolName}`,
                             streaming: false,
                             toolCall: {
                                 name: toolCall.function.name,
-                                arguments: toolArgs
+                                arguments: toolArgs,
                             },
                             toolDisplay,
                             toolCallId: toolCall.id, // Store tool call ID for later update
-                            toolPending: true // Mark as pending execution
-                        }]);
+                            toolPending: true, // Mark as pending execution
+                        },
+                    ]);
                 }
                 // Filter tools that need confirmation (not in always-approved list OR session-approved list)
                 const toolsNeedingConfirmation = [];
                 const autoApprovedTools = [];
                 for (const toolCall of receivedToolCalls) {
                     // Check both global approved list and session-approved list
-                    if (isToolAutoApproved(toolCall.function.name) || sessionApprovedTools.has(toolCall.function.name)) {
+                    if (isToolAutoApproved(toolCall.function.name) ||
+                        sessionApprovedTools.has(toolCall.function.name)) {
                         autoApprovedTools.push(toolCall);
                     }
                     else {
@@ -267,105 +269,42 @@ export async function handleConversationWithTools(options) {
                 }
                 else if (toolsNeedingConfirmation.length > 0) {
                     const firstTool = toolsNeedingConfirmation[0]; // Safe: length > 0 guarantees this exists
-                    // Check if we should use DIFF+APPLY for filesystem tools
-                    let usedDiffApply = false;
-                    if (vscodeConnection.isConnected()) {
-                        // Try to use DIFF+APPLY for filesystem_create or filesystem_edit
-                        if (firstTool.function.name === 'filesystem-create' || firstTool.function.name === 'filesystem-edit') {
-                            try {
-                                const args = JSON.parse(firstTool.function.arguments);
-                                if (firstTool.function.name === 'filesystem-create') {
-                                    // For create, show diff with empty old content
-                                    const confirmation = await vscodeConnection.requestDiffApply(args.filePath, '', // Empty old content for new file
-                                    args.content);
-                                    if (confirmation === 'reject') {
-                                        // Remove pending tool messages
-                                        setMessages(prev => prev.filter(msg => !msg.toolPending));
-                                        setMessages(prev => [...prev, {
-                                                role: 'assistant',
-                                                content: 'Tool call rejected, session ended',
-                                                streaming: false
-                                            }]);
-                                        if (options.setIsStreaming) {
-                                            options.setIsStreaming(false);
-                                        }
-                                        encoder.free();
-                                        return;
-                                    }
-                                    if (confirmation === 'approve_always') {
-                                        addMultipleToAlwaysApproved([firstTool.function.name]);
-                                        sessionApprovedTools.add(firstTool.function.name);
-                                    }
-                                    approvedTools.push(...toolsNeedingConfirmation);
-                                    usedDiffApply = true;
-                                }
-                                else if (firstTool.function.name === 'filesystem-edit') {
-                                    // For edit, read the file first to get old content
-                                    const fileContent = await filesystemService.getFileContent(args.filePath, args.startLine, args.endLine);
-                                    const confirmation = await vscodeConnection.requestDiffApply(args.filePath, fileContent.content, args.newContent);
-                                    if (confirmation === 'reject') {
-                                        // Remove pending tool messages
-                                        setMessages(prev => prev.filter(msg => !msg.toolPending));
-                                        setMessages(prev => [...prev, {
-                                                role: 'assistant',
-                                                content: 'Tool call rejected, session ended',
-                                                streaming: false
-                                            }]);
-                                        if (options.setIsStreaming) {
-                                            options.setIsStreaming(false);
-                                        }
-                                        encoder.free();
-                                        return;
-                                    }
-                                    if (confirmation === 'approve_always') {
-                                        addMultipleToAlwaysApproved([firstTool.function.name]);
-                                        sessionApprovedTools.add(firstTool.function.name);
-                                    }
-                                    approvedTools.push(...toolsNeedingConfirmation);
-                                    usedDiffApply = true;
-                                }
-                            }
-                            catch (error) {
-                                // If DIFF+APPLY fails, fall back to regular confirmation
-                                console.error('Failed to use DIFF+APPLY:', error);
-                            }
+                    // Use regular CLI confirmation
+                    // Pass all tools for proper display in confirmation UI
+                    const allTools = toolsNeedingConfirmation.length > 1
+                        ? toolsNeedingConfirmation
+                        : undefined;
+                    // Use first tool for confirmation UI, but apply result to all
+                    const confirmation = await requestToolConfirmation(firstTool, undefined, allTools);
+                    if (confirmation === 'reject') {
+                        // Remove pending tool messages
+                        setMessages(prev => prev.filter(msg => !msg.toolPending));
+                        // User rejected - end conversation
+                        setMessages(prev => [
+                            ...prev,
+                            {
+                                role: 'assistant',
+                                content: 'Tool call rejected, session ended',
+                                streaming: false,
+                            },
+                        ]);
+                        // End streaming immediately
+                        if (options.setIsStreaming) {
+                            options.setIsStreaming(false);
                         }
+                        encoder.free();
+                        return; // Exit the conversation loop
                     }
-                    // If we didn't use DIFF+APPLY, fall back to regular CLI confirmation
-                    if (!usedDiffApply) {
-                        // Pass all tools for proper display in confirmation UI
-                        const allTools = toolsNeedingConfirmation.length > 1
-                            ? toolsNeedingConfirmation
-                            : undefined;
-                        // Use first tool for confirmation UI, but apply result to all
-                        const confirmation = await requestToolConfirmation(firstTool, undefined, allTools);
-                        if (confirmation === 'reject') {
-                            // Remove pending tool messages
-                            setMessages(prev => prev.filter(msg => !msg.toolPending));
-                            // User rejected - end conversation
-                            setMessages(prev => [...prev, {
-                                    role: 'assistant',
-                                    content: 'Tool call rejected, session ended',
-                                    streaming: false
-                                }]);
-                            // End streaming immediately
-                            if (options.setIsStreaming) {
-                                options.setIsStreaming(false);
-                            }
-                            encoder.free();
-                            return; // Exit the conversation loop
-                        }
-                        // If approved_always, add ALL these tools to both global and session-approved sets
-                        if (confirmation === 'approve_always') {
-                            const toolNamesToAdd = toolsNeedingConfirmation.map(t => t.function.name);
-                            // Add to global state (async, for future sessions)
-                            addMultipleToAlwaysApproved(toolNamesToAdd);
-                            // Add to local session set (sync, for this conversation)
-                            toolNamesToAdd.forEach(name => sessionApprovedTools.add(name));
-                        }
-                        // Add all tools to approved list
-                        approvedTools.push(...toolsNeedingConfirmation);
+                    // If approved_always, add ALL these tools to both global and session-approved sets
+                    if (confirmation === 'approve_always') {
+                        const toolNamesToAdd = toolsNeedingConfirmation.map(t => t.function.name);
+                        // Add to global state (async, for future sessions)
+                        addMultipleToAlwaysApproved(toolNamesToAdd);
+                        // Add to local session set (sync, for this conversation)
+                        toolNamesToAdd.forEach(name => sessionApprovedTools.add(name));
                     }
+                    // Add all tools to approved list
+                    approvedTools.push(...toolsNeedingConfirmation);
                 }
                 // Execute approved tools
                 const toolResults = await executeToolCalls(approvedTools);
@@ -390,7 +329,9 @@ export async function handleConversationWithTools(options) {
                         const statusText = isError ? `\n  └─ ${result.content}` : '';
                         // Check if this is an edit tool with diff data
                         let editDiffData;
-                        if ((toolCall.function.name === 'filesystem-edit' || toolCall.function.name === 'filesystem-edit_search') && !isError) {
+                        if ((toolCall.function.name === 'filesystem-edit' ||
+                            toolCall.function.name === 'filesystem-edit_search') &&
+                            !isError) {
                             try {
                                 const resultData = JSON.parse(result.content);
                                 if (resultData.oldContent && resultData.newContent) {
@@ -400,7 +341,7 @@ export async function handleConversationWithTools(options) {
                                         filename: JSON.parse(toolCall.function.arguments).filePath,
                                         completeOldContent: resultData.completeOldContent,
                                         completeNewContent: resultData.completeNewContent,
-                                        contextStartLine: resultData.contextStartLine
+                                        contextStartLine: resultData.contextStartLine,
                                     };
                                 }
                             }
@@ -418,7 +359,7 @@ export async function handleConversationWithTools(options) {
                                         stdout: resultData.stdout || '',
                                         stderr: resultData.stderr || '',
                                         exitCode: resultData.exitCode || 0,
-                                        command: resultData.command
+                                        command: resultData.command,
                                     };
                                 }
                             }
@@ -433,15 +374,19 @@ export async function handleConversationWithTools(options) {
                                     ...msg,
                                     content: `${statusIcon} ${toolCall.function.name}${statusText}`,
                                     toolPending: false,
-                                    toolCall: editDiffData ? {
-                                        name: toolCall.function.name,
-                                        arguments: editDiffData
-                                    } : terminalResultData ? {
-                                        name: toolCall.function.name,
-                                        arguments: terminalResultData
-                                    } : msg.toolCall,
+                                    toolCall: editDiffData
+                                        ? {
+                                            name: toolCall.function.name,
+                                            arguments: editDiffData,
+                                        }
+                                        : terminalResultData
+                                            ? {
+                                                name: toolCall.function.name,
+                                                arguments: terminalResultData,
+                                            }
+                                            : msg.toolCall,
                                     // Store tool result for preview rendering
-                                    toolResult: !isError ? result.content : undefined
+                                    toolResult: !isError ? result.content : undefined,
                                 };
                             }
                             return msg;
@@ -461,8 +406,8 @@ export async function handleConversationWithTools(options) {
                             role: 'assistant',
                             content: '',
                             streaming: false,
-                            showTodoTree: true
-                        }
+                            showTodoTree: true,
+                        },
                     ]);
                 }
                 // Check if there are pending user messages to insert
@@ -474,17 +419,20 @@ export async function handleConversationWithTools(options) {
                         // Combine multiple pending messages into one
                         const combinedMessage = pendingMessages.join('\n\n');
                         // Add user message to UI
-                        const userMessage = { role: 'user', content: combinedMessage };
+                        const userMessage = {
+                            role: 'user',
+                            content: combinedMessage,
+                        };
                         setMessages(prev => [...prev, userMessage]);
                         // Add user message to conversation history
                         conversationMessages.push({
                             role: 'user',
-                            content: combinedMessage
+                            content: combinedMessage,
                         });
                         // Save user message
                         saveMessage({
                             role: 'user',
-                            content: combinedMessage
+                            content: combinedMessage,
                         }).catch(error => {
                             console.error('Failed to save pending user message:', error);
                         });
@@ -499,13 +447,13 @@ export async function handleConversationWithTools(options) {
                     role: 'assistant',
                     content: streamedContent.trim(),
                     streaming: false,
-                    discontinued: controller.signal.aborted
+                    discontinued: controller.signal.aborted,
                 };
                 setMessages(prev => [...prev, finalAssistantMessage]);
                 // Add to conversation history and save
                 const assistantMessage = {
                     role: 'assistant',
-                    content: streamedContent.trim()
+                    content: streamedContent.trim(),
                 };
                 conversationMessages.push(assistantMessage);
                 saveMessage(assistantMessage).catch(error => {

package/dist/hooks/useKeyboardInput.js

@@ -26,6 +26,15 @@ export function useKeyboardInput(options) {
     useInput((input, key) => {
         if (disabled)
             return;
+        // Shift+Tab - Toggle YOLO mode
+        if (key.shift && key.tab) {
+            executeCommand('yolo').then(result => {
+                if (onCommand) {
+                    onCommand('yolo', result);
+                }
+            });
+            return;
+        }
         // Handle escape key for double-ESC history navigation
         if (key.escape) {
             // Close file picker if open

package/dist/mcp/filesystem.d.ts

@@ -131,8 +131,6 @@ export declare class FilesystemMCPService {
         totalLines: number;
         structureAnalysis?: StructureAnalysis;
         diagnostics?: Diagnostic[];
-        completeOldContent?: string;
-        completeNewContent?: string;
     }>;
     /**
      * Edit a file by replacing lines within a specified range
@@ -158,8 +156,6 @@ export declare class FilesystemMCPService {
         linesModified: number;
         structureAnalysis?: StructureAnalysis;
         diagnostics?: Diagnostic[];
-        completeOldContent?: string;
-        completeNewContent?: string;
     }>;
     /**
      * Resolve path relative to base path and normalize it