agent-sh/agentsys

What it solves

AgentSys addresses the complexity of the software development lifecycle that exists beyond simple code generation. While AI models can write code, managing task selection, branch management, code review, CI/CD, and artifact cleanup remains difficult. AgentSys provides a modular runtime and orchestration system to automate these processes through structured pipelines and specialized agents.

How it works

AgentSys acts as a marketplace and installer for a collection of plugins, agents, and skills. It orchestrates these components into phase-gated pipelines where each agent has a specific responsibility, model assignment, and defined inputs/outputs.

To optimize cost and performance, it uses a hybrid approach: deterministic tools (regex, AST analysis) handle detection, while LLMs are reserved for high-level judgment and synthesis. It supports persistent state across sessions and integrates with multiple AI coding platforms including Claude Code, OpenCode, Codex CLI, Cursor, and Kiro.

Who it’s for

Developers and teams using AI coding assistants who want to move from manual prompting to automated, reliable software engineering workflows.

Highlights

• Comprehensive Command Suite: Includes tools for task discovery (/next-task), quality gating (/prepare-delivery), and durable negative memory (/banthis).
• Agnix Integration: Features a dedicated linter for AI agent configurations with 457 rules to catch errors in skills and prompts before they fail.
• Model Agnostic Efficiency: Benchmarks show that its structured pipelines allow smaller models (like Claude Sonnet) to achieve quality comparable to larger models (like Claude Opus) at a significantly lower cost.
• Modular Ecosystem: Comprises 24 plugins, 50 agents, and 44 skills distributed as standalone repositories.