CodeSkeleton AI

Focus Extractor & Architectural Outliner for LLMs

Developed by KiranLabs

Stop AI models from skipping lines and hallucinating on large codebases. Supports both Function-Level Focus and File / Folder Architectural Skeletons.

🌟 Why CodeSkeleton AI?

When pasting code into ChatGPT, Claude 3.5, Gemini 1.5 Pro, or DeepSeek, models suffer from attention degradation on long files:

• They skim through hundreds of lines of boilerplate.
• They get lazy and ignore critical lines of logic.
• They answer with // ... existing code ... or hallucinate non-existent variables.

CodeSkeleton AI solves this on both the Function and File levels:

• 🎯 Function-Level Focus: Place your cursor inside any function (or highlight code) -> CodeSkeleton automatically detects the enclosing function boundaries, keeps that target 100% intact, and collapses all other functions in the file into signatures ({ /* ... omitted ... */ }).
• 📄 File-Level Skeleton: Right-click any file in the Editor or Explorer sidebar to instantly generate a complete structural map of imports, exports, classes, and method signatures with all bodies collapsed.
• 📁 Folder / Multi-File Skeletons: Right-click any folder or multiple files in Explorer to generate a compact architectural blueprint of your entire module or workspace.
• 🧠 Zero Attention Fatigue: Reduces token load by 70%–90%, keeping the LLM's attention sharp and focused on your target code.

🚀 How to Use

1. Function Level (Inside the Editor)

1. Place your cursor anywhere inside a function (or highlight a block of code).
2. Right-click and choose "CodeSkeleton AI: Focus on Function / Selection (Collapse File)".
3. CodeSkeleton AI auto-detects the enclosing function, preserves it with 100% full implementation, collapses all other functions, and copies the result with an AI directive to your clipboard!

2. File Level (From Editor or Explorer)

1. Right-click any file in the Explorer sidebar or in an active editor.
2. Select "CodeSkeleton AI: Copy File Skeleton (Collapse All Functions)".
3. All function bodies are collapsed into signatures, preserving imports, classes, and types.

3. Folder / Workspace Level (From Explorer)

1. Right-click any folder (or select multiple files) in VS Code's Explorer sidebar.
2. Select "CodeSkeleton AI: Copy Folder / Selected Files Skeleton".
3. It recursively gathers all code files (skipping binaries & .git/node_modules) and generates a unified multi-file architectural skeleton!

4. Export to File

• Choose "CodeSkeleton AI: Export Skeleton to .txt File" to save the generated skeleton directly to disk.

📄 Output Preview

Function Focus Example:

================================================================================
CODESKELETON AI: PaymentService.js
Original Lines: 284 | Skeletons Collapsed: 8
Estimated Tokens: ~340 (Saved ~1,850 tokens / 84% reduction)
================================================================================
🎯 FOCUS TARGET: 'processRefund' (Lines 142-178 - 100% full implementation shown)
💡 AI DIRECTIVE: Please read EVERY SINGLE LINE of the Focus Target carefully.
================================================================================

   1 | import { Database } from '../db';
   2 | import { PaymentGateway } from './gateway';
   3 | 
   4 | export class PaymentService {
   5 |   constructor(private db: Database) { /* ... omitted ... */ }
   6 |   async validateCard(token: string): Promise<boolean> { /* ... omitted ... */ }
   7 | 
>142 |   async processRefund(transactionId: string) {
>143 |     const tx = await this.db.find(transactionId);
>144 |     if (!tx || tx.status !== 'completed') {
>145 |       throw new Error('Invalid transaction status');
>146 |     }
>147 |     return await this.gateway.refund(tx.id, tx.amount);
>148 |   }
 149 | 
 150 |   async sendReceipt(email: string): Promise<void> { /* ... omitted ... */ }
 151 | }

⚙️ Settings

Configure preferences in VS Code Settings (Ctrl + , -> search CodeSkeleton AI):

📦 About KiranLabs

KiranLabs builds developer tooling to make programming with AI fast, seamless, and reliable.

• CodePacker AI
• CodeSkeleton AI
• PatchBack AI

📄 License

MIT License. Free to use, modify, and distribute.