ShunsukeHayashi

ShunsukeHayashi / agent-skill-bus

Public

Self-improving task orchestration framework for AI agent systems. DAG-based task queue + automatic skill quality monitoring + external change detection. Framework-agnostic (OpenClaw, Claude Code, Codex, LangGraph, CrewAI).

www.miyabi-ai.jp agent-framework ai-agents claude-code codex dag-scheduler
19
3
100% credibility
Found Mar 19, 2026 at 19 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
JavaScript
AI Summary

Agent Skill Bus is a simple organizer for AI agent tasks that queues jobs safely with order and locks, tracks performance, detects issues, and triggers automatic improvements.

How It Works

1
🔍 Discover the Helper

You find a simple toolkit that keeps your AI assistants organized and getting better on their own.

2
🛠️ Set It Up Quickly

You prepare the organizer in your work folder with one easy action, creating spaces for tasks and notes.

3
📝 Add Your First Task

You list a job for your AI helper, like 'fix this issue', marking what files it touches and if it waits for others.

4
🚀 Send Tasks to Helpers

Ready jobs go to your AI assistants one by one, safely without overlaps or missing steps.

5
📊 Track Results and Health

After each job, you note if it succeeded and how well, and the system spots any weak spots.

6
🔄 Watch for Changes

It notices outside updates that might affect your helpers and suggests fixes automatically.

✅ Smarter Assistants

Your AI team runs smoothly, learns from mistakes, and stays reliable as it improves itself over time.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 19 to 19 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 agent-skill-bus?

Agent Skill Bus is a JavaScript framework-agnostic toolkit for self-improving AI agent systems, handling task orchestration via a DAG-based queue, automatic skill quality monitoring, and external change detection. It prevents common prod pitfalls like degrading skills from API updates, colliding agent tasks on shared files, unordered dependencies, and repeating failures without feedback loops. Users get a simple CLI—npx agent-skill-bus init, enqueue tasks, dispatch ready ones, record-run scores—to integrate with any agent setup using plain JSONL files.

Why is it gaining traction?

Unlike LangGraph or CrewAI that focus on execution, this adds operational health: file locking stops data corruption, priority routing and deduping streamline workflows, and self-improving loops auto-flag drifts or failures for repair. The hook is dead-simple integration—drop CLI calls into Claude Code, Codex, or CrewAI agents—and zero deps or DBs, making it a lightweight bus for reliable, learning agents. Early adopters note 57% fewer failures in prod.

Who should use this?

AI engineers coordinating multiple agents in production, like teams running CrewAI or LangGraph swarms needing DAG task ordering and collision-free file edits. Claude Code or Codex users wanting automatic skill monitoring to catch auth expires or model drifts. Small ops teams handling human-triggered, cron, or webhook tasks without heavy infra.

Verdict

Worth prototyping for self-improving agent GitHub setups if you're tired of brittle agent ops—solid CLI, tests, and docs punch above 19 stars and 1.0% credibility. Still early; scale cautiously until more battle reports emerge.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.