context-trimmer

A zero-overhead AST-based code skeletonizer and token compressor for LLMs and AI coding agents.

---

The Problem

Passing whole repositories or massive source files into LLM prompts quickly causes:

1. Context Window Exhaustion: Thousands of tokens burned on implementation trivia (loops, math, internal boilerplate).
2. Attention Degradation (Lost-in-the-middle): Agents struggle to identify relevant interfaces and function signatures buried inside dense files.

The Solution

context-trimmer parses your codebase into an AST and strips function and method bodies, replacing them with concise placeholders while strictly preserving:

• import statements and dependencies
• type, interface, and enum definitions
• Function, method, and constructor signatures with parameter and return types
• JSDoc comments and Python docstrings
• Class hierarchy and visibility modifiers (public, private, readonly)

---

Quick Benchmarks

On typical production TypeScript and Python codebases:

• Token reduction: 55% to 78%
• Speed: Over 10,000 LOC parsed and trimmed in under 120ms (zero native C/node-gyp compilation).

---

Installation & Usage

Direct execution with npx:

# Print skeleton of a file or directory
npx context-trimmer ./src

# View token reduction stats
npx context-trimmer ./src --stats

# Export trimmed context to a single file for feeding to LLMs
npx context-trimmer ./src --out context.txt

---

Example

Before:

export function processTransaction(id: string, amount: number): Receipt {
  const tax = amount * 0.21;
  const total = amount + tax;
  const receipt = generateDatabaseRecord(id, total);
  notifyAccounting(receipt);
  return receipt;
}

After:

export function processTransaction(id: string, amount: number): Receipt { /* ... */ }

---

License

MIT © luckdevx