anjiacm/agent-kit
Analytics & monitoring이 페이지는 AI로 자동 번역되었습니다. 기술적인 세부 사항은 원문을 확인하세요. 원문 보기
Production-ready Agent Runtime for Claude Code — Dashboard visualization, Team mode, Plugin daemons, Memory system, and one-command setup
개요
Agent Kit is the for Claude Code — giving it persistence, observability, and multi-process coordination. In short: It has tool calling, bash execution, file editing, hooks, team spawning — everything needed for powerful single-session work. It doesn't have a scheduler, a persistent memory system, a task queue, state recovery, or a lifecycle manager. To build a real autonomous agent — one that runs 24/7, survives context compression, coordinates multiple workers, and improves over time — you need an entire runtime layer on top. Click to watch: Create → Launch → Team Spawn → Dashboard → Cross-Process Communication — full loop in 2 minutes Claude Code provides the raw capabilities. Agent Kit adds the 6 things needed to turn a conversation into a system: , building this from scratch takes 2-4 weeks and dozens of pitfalls (documented in proven patterns). : one command, 5 minutes, production-ready. These agents run daily in production, built entirely with this framework:
README
원문 README의 일부입니다. 원문 보기.
Agent Kit
Agent Kit is the Agent Runtime for Claude Code — giving it persistence, observability, and multi-process coordination. In short: the body for the brain.
Claude Code is the most capable AI coding CLI available today. It has tool calling, bash execution, file editing, hooks, team spawning — everything needed for powerful single-session work.
But it’s request-response by design. It doesn’t have a scheduler, a persistent memory system, a task queue, state recovery, or a lifecycle manager. To build a real autonomous agent — one that runs 24/7, survives context compression, coordinates multiple workers, and improves over time — you need an entire runtime layer on top.
Agent Kit is that runtime layer.
bash create-agent.sh
Demo Video (2 min)
Click to watch: Create → Launch → Team Spawn → Dashboard → Cross-Process Communication — full loop in 2 minutes
What Agent Kit Actually Builds On Top of Claude Code
Claude Code provides the raw capabilities. Agent Kit adds the 6 things needed to turn a conversation into a system:
| What’s Missing | Why It Matters | What Agent Kit Provides |
|---|---|---|
| Agent Loop | No event loop = only acts when you speak | Plugin daemons (nohup) run independent monitoring cycles 24/7 |
| Memory | Context compression = knowledge lost | Per-entity Markdown files + cross-entity knowledge base, persisted across sessions |
| Task Queue | No way to queue and dispatch work | Message queue (REST API) + background polling + multi-channel input (terminal, Dashboard, Feishu) |
| State Management | Workers die silently after compression | Centralized State Protocol: heartbeat registry + state ledger + precise per-worker recovery in <5s |
| Scheduler | No time-based or event-based triggering | Hook-driven lifecycle (session start/stop/idle) + cron heartbeats + daemon task scheduling |
| Observability | Terminal output is a black box | Real-time Dashboard: pixel-art workers, live terminal tabs, cancel buttons, status panels |
Without Agent Kit, building this from scratch takes 2-4 weeks and dozens of pitfalls (documented in proven patterns).
With Agent Kit: one command, 5 minutes, production-ready.
Production-Proven Agents
These agents run daily in production, built entirely with this framework:
Server Maintenance Agent
12 servers. 4 workers. 10+ monitoring tasks. Fully autonomous.
- Manages 7 production servers across 3 countries via SSH
- 4 parallel Workers handle health checks, deployments, log analysis, Nginx/SSL management
- Cloudflare WAF monitoring catches carding bots, blocks malicious ASNs
추천 도구
다른 키워드를 입력하거나 필터를 제거해 보세요.
설치
npx skillfish add anjiacm/agent-kit