Skillget

choruskube

workflow
Guvenlik Denetimi
Basarisiz
Health Uyari
  • License — License: AGPL-3.0
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Low visibility — Only 5 GitHub stars
Code Basarisiz
  • rm -rf — Recursive force deletion command in agent-images/choruskube-dind/test/test-build-preload.sh
  • rm -rf — Recursive force deletion command in agent-images/choruskube-dind/test/test-dind-entrypoint.sh
  • exec() — Shell command execution in agent-images/claude-code/harvest-junit.js
  • fs module — File system access in agent-images/claude-code/harvest-junit.js
  • rm -rf — Recursive force deletion command in agent-images/claude-code/test/test-check-prs.sh
Permissions Gecti
  • Permissions — No dangerous permissions requested

Bu listing icin henuz AI raporu yok.

SUMMARY

AI-agent workflow orchestrator for Kubernetes and Docker — single-tenant OSS core

README.md

ChorusKube

An AI-agent workflow orchestrator for Kubernetes and Docker.

Giving an AI coding agent free rein over your codebase is a leap of faith.
ChorusKube lets you define work as a directed graph — AI tasks, scripts, and
human-approval gates — and runs each node as an isolated container, so agents
do the work while you keep control. AI nodes run Claude Code
against a target git repo; a Temporal-backed orchestrator drives the graph and
a Worker owns the container lifecycle.

This is the open-source core: single-tenant, no external identity provider —
the API is open and every request is scoped to a seeded "system" organization.

Why ChorusKube?

  • Workflow as a DAG — model spec → plan → implement → test → review → merge
    as a graph with mandatory gates. Conditional routing and revision loops included.
  • Container isolation — every node runs in its own ephemeral container with a
    unique per-execution secret; agents get autonomy in a sandbox, not your cluster.
  • Human-in-the-loop — pause at review gates, inspect predecessor outputs and
    artifacts, then approve, reject, or request changes that feed the next iteration.
  • Live monitoring — watch the graph advance node-by-node over WebSocket; the
    UI subscribes to events rather than polling.

Node types

Type What it does
ai An AI agent (Claude Code) runs autonomously in a container
script A shell command runs in a container
human The run pauses for approval via the dashboard
both AI produces output, then a human reviews before continuing

Components

  • api-server — Spring Boot (Java 25). Source of truth for run state; resolves
    per-run credentials and records Worker-reported execution state.
  • orchestrator — Go + Temporal. Drives the graph; dispatches each node as a
    Temporal activity for a Worker to execute.
  • worker — Go + Temporal. Runs executors (Docker, Kubernetes) and the
    agent callback server; receives work from Fleets.
  • web-ui — React + Vite. Graph visualization, live monitoring, approval gates.
  • agent images — claude-code (the AI agent) and choruskube-dev (a fuller
    dev toolchain on top).

Getting started

Needs Docker + Docker Compose v2. See QUICKSTART.md to boot
the stack, ARCHITECTURE.md for how it fits together, and
PRIVACY.md for the anonymous opt-out telemetry.

License

AGPL-3.0.

Yorumlar (0)

Sonuc bulunamadi