A modular runtime and orchestration system for AI agents that automates software development workflows from task discovery to deployment.
AgentSys is a modular runtime and orchestration system designed to automate the entire software development lifecycle using specialized AI agents. It handles tasks beyond code generation—such as task selection, branch management, code review, artifact cleanup, CI, PR comments, and deployment—through structured, phase-gated pipelines with persistent state that survives session interruptions.
Developers and teams using AI coding assistants (like Claude Code, OpenCode, Codex CLI, Cursor, or Kiro) who want to automate and orchestrate the complete development workflow, from task discovery to deployment, with minimal manual oversight.
Developers choose AgentSys for its comprehensive, automated pipeline that enforces quality gates and reduces token usage by up to 77% through deterministic tools for detection and single LLM calls for judgment. Its modular architecture with 47 specialized agents and persistent state allows for reliable, multi-session automation without babysitting.
AI writes code. This automates everything else · 20 plugins · 49 agents · 41 skills · for Claude Code, OpenCode, Codex, Cursor, Kiro.
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.
With 19 plugins, 47 agents, and 40 skills, AgentSys allows teams to build customized pipelines for software development, enabling standalone use of commands like `/deslop` or integrated workflows via `/next-task`.
By using deterministic tools like regex and AST analysis for detection and single LLM calls for judgment, AgentSys reduces token usage by up to 77%, as demonstrated in the `/drift-detect` benchmark.
It runs on Claude Code, OpenCode, Codex CLI, Cursor, and Kiro with consistent behavior, and tools like `/agnix` extend support to more AI assistants, ensuring broad usability.
Commands like `/prepare-delivery` enforce pre-ship checks including deslopping, multi-agent review, and documentation sync, preventing skipped steps and ensuring code quality.
Installation requires Node.js 18+, Git, and additional CLIs like agnix and agent-analyzer, with platform-specific configurations for AI tools, which can be a barrier to entry and maintenance.
The system is optimized for Claude models (Opus, Sonnet, Haiku) with specific model assignments, and while it supports other platforms, core functionality and token savings are tied to this ecosystem, limiting flexibility.
Understanding the architecture of 47 specialized agents, 40 skills, and how commands compose requires significant time investment, as highlighted in the design philosophy section about research-backed prompt engineering.