Blog / · GH-Proxy Team

Fast-Jev-Compaction: Precision Context Pruning and Decision Compaction for Claude Code

#Open Source #FastJevCompaction #ClaudeCode #Context Window #GitHub Proxy #DevTools
Table of Contents4sections

Fast-Jev-Compaction: Precision Context Pruning and Decision Compaction for Claude Code

When conducting complex, hours-long software refactors with Claude Code, context exhaustion is an ever-present friction point. Standard compaction routines delegate history pruning to a generic LLM summarizer. However, summaries are fundamentally lossy: exact stack traces, configuration hashes, file paths, and edge constraints frequently evaporate from memory.

fast-jev-compaction introduces an elegant, deterministic alternative: never rewrite historical dialogue; instead, leverage sub-second decision scoring to surgically prune obsolete tool outputs while keeping all conversational text completely verbatim. Rising to 7,000+ Stars on GitHub with 2,880+ Stars added this week, it is becoming a must-have extension for power users.

🚀 Accelerated Setup via GH-Proxy

Download the package and Claude Code hooks without international network latency via GH-Proxy:

👉 Direct GH-Proxy Acceleration for Fast-Jev-Compaction
(Alternative URL: https://gh-proxy.com/https://github.com/tamaratran/fast-jev-compaction)

Installation:

git clone https://gh-proxy.com/https://github.com/tamaratran/fast-jev-compaction.git
cd fast-jev-compaction
npm install
npm run build

🔍 How Lossless Compaction Works

Rather than rewriting historical dialogue, Fast-Jev-Compaction acts as an intelligent surgical scalpel:

  1. Single-Pass Evaluation: Evaluates all previous tool calls, bash stdout traces, and filesystem searches against the overarching task state in a single rapid request.
  2. Relevance Scoring: Marks intermediate, superseded outputs (such as stale directory listings, resolved compiler diagnostics, and redundant test outputs) as disposable.
  3. Verbatim Preservation: Safely drops or truncates unneeded tool payload data while keeping every developer prompt, architectural decision, and active diff intact and strictly ordered.

🛠 Quick Configuration

# Install as Claude Code plugin
claude plugin install fast-jev-compaction

# Or import into custom agent loops
npm install fast-jev-compaction

import { compactHistoryWithJev } from 'fast-jev-compaction';

const optimizedContext = await compactHistoryWithJev(conversationHistory, {
  maxContextTokens: 64000,
  keepRecentToolCalls: 5
});

console.log("Context window pruned cleanly without semantic loss.");

🎯 Key Benefits

  • Zero Loss of Critical Exact Strings: Eliminates subtle bugs where file names or parameter flags disappear during summarization.
  • Cheaper and Faster: Skips expensive multi-second token generation rounds during compaction triggers.
  • Unbroken Coding Continuity: Enables sustained multi-hour coding sessions without agent drift.

💬 Comments & Discussion