mega-edo

The skill OS for Codex, Claude Code, and Gemini CLI. One pool, one router, one feedback loop โ€” across all three hosts. Per-turn semantic top-K with dynamic context sizing, session-end self-evaluation, and evidence-blended re-ranking that gets better the more you use it.

14
0
89% credibility
Found May 22, 2026 at 14 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

mega-tron is a local skill management layer that makes AI coding assistants (Codex, Claude Code, and Gemini CLI) smarter about which skills to use. Instead of loading hundreds of skills into every conversation (wasting tokens and money), it analyzes what you're actually asking and only loads the relevant onesโ€”cutting token usage by 10-30x while improving answer quality. It also unifies your skill library across all three assistants so edits sync everywhere, and it learns from outcomes to improve future skill selection. Everything runs on your machine for privacy. Built by the MEGA Code platform with Apache 2.0 license.

How It Works

1
๐Ÿ’ญ You notice your AI assistant is slow and expensive

You type a simple question and wait forever. Your AI coding tool loads hundreds of skills you never asked for, eating up your budget.

2
๐Ÿ” You discover mega-tron

A friend or search tells you about a tool that makes your AI assistant smarter about which skills to use.

3
โšก You install it in two commands

You run a simple installer, and mega-tron automatically connects to your AI coding tools (Codex, Claude, and Gemini).

4
โœจ Everything suddenly works better

The next time you ask your AI assistant something, it only loads the skills that actually matter for your question. Tokens drop 10-30x.

5
Your skills stay in sync everywhere
๐Ÿ“Š
You open the dashboard

A beautiful web page shows you which skills are actually helping and which ones broke silently.

๐Ÿค–
You keep working normally

The system learns quietly in the background, getting smarter with every session.

๐ŸŽ‰ Your AI assistant gets smarter over time

Skills that help rise to the top. Broken skills retire themselves. You saved tokens and got better answers.

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 14 to 14 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is mega-tron?

Mega-tron is a local skill routing layer for AI coding assistants. It sits above Codex, Claude Code, and Gemini CLI and fixes three compounding problems: native hosts dump your entire skill catalog into every prompt regardless of relevance (wasting tokens), skills drift out of sync across hosts (editing one doesn't update the others), and you have no visibility into which skills actually helped. Mega-tron embeds your prompt, ranks skills semantically against it, and ships only the relevant ones - roughly 600 tokens per turn whether you have 30 skills or 500. It also maintains a unified skill pool with cross-host symlinks and records HELPFUL/HARMFUL verdicts that feed back into future routing decisions.

Why is it gaining traction?

The token savings are dramatic and immediate. Benchmarks show 11-187x fewer tokens shipped per turn compared to native hosts, with higher coverage. But the real hook is the feedback loop: the router learns from how you actually work. Skills that break against library updates get auto-archived; skills that consistently help get ranked higher across all hosts. The dashboard surfaces this entire verdict economy so you're not guessing why "the answer felt weird." Everything runs locally with no API calls, so it works offline and costs nothing per turn.

Who should use this?

Developers running multiple AI coding assistants with large skill collections. If you're using Codex, Claude Code, or Gemini CLI and have accumulated 20+ skills, you're likely shipping thousands of tokens per turn for skills that don't apply to your current task. Teams maintaining shared skill libraries across different agents will benefit most from the unified pool and cross-host verdict economy.

Verdict

This is a clever solution to a real problem with solid benchmarks backing it up. The architecture is well-thought-out and the self-improving feedback loop is genuinely novel. However, with 14 stars and a credibility score of 0.9%, it's early-stage software from an unknown author. The documentation is thorough and the install process is straightforward, but you'd be an early adopter. Worth trying if token costs or skill management across hosts are burning you - just don't bet production workflows on it yet.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.