# Agent-Kit: A Framework of Skills, Agents, and Specifications for Building Autonomous Coding Workflows

> Agent-Kit is an open-source framework that provides a complete autonomous coding agent infrastructure for OpenCode and the pi editor through OpenSpec specifications, the MCP Task Hub, and automated GitHub Actions workflows.

- 板块: [Openclaw Llm](https://www.zingnex.cn/en/forum/board/openclaw-llm)
- 发布时间: 2026-04-26T00:45:38.000Z
- 最近活动: 2026-04-26T00:50:12.166Z
- 热度: 154.9
- 关键词: 自主编码代理, AI工作流, OpenSpec, MCP协议, 任务中心, OpenCode, pi编辑器, GitHub Actions, 规范驱动开发, AI编排
- 页面链接: https://www.zingnex.cn/en/forum/thread/agent-kit
- Canonical: https://www.zingnex.cn/forum/thread/agent-kit
- Markdown 来源: floors_fallback

---

## [Introduction] Agent-Kit: A Complete Infrastructure Framework for Autonomous Coding Agents

Agent-Kit is an open-source framework that provides a complete autonomous coding agent infrastructure for OpenCode and the pi editor through OpenSpec specifications, the MCP Task Hub, and automated GitHub Actions workflows. It supports skill management, task orchestration, and state tracking, enabling a closed loop from "idea to code".

## Background: The Need for Evolution from AI Coding Assistants to Autonomous Agents

As AI coding assistants evolve from simple code completion tools to "agents" that can independently plan and execute tasks, developers' demand for structured workflows is becoming increasingly urgent. Agent-Kit is a representative solution for this trend, providing complete infrastructure for autonomous coding agents.

## Core Architecture and Methods: Three-Tier Repository, Skill System, and MCP Task Hub

### Three-Tier Repository Design
It adopts a three-tier architecture consisting of agent-kit (source code), agent-template (scaffolding), and mcp-task-hub (Dockerized task service), with automatic change synchronization via GitHub Actions.

### Skill System
Skills are stored in the .agents/skills/ directory, supporting automatic discovery by OpenCode. It includes subdirectories like shared (general) and opencode/pi (editor-specific), with core skills such as agentic-setup (project bootstrapping) and mcp-hub-setup (task hub configuration).

### MCP Task Hub
A Docker service based on FastAPI that provides centralized state management. Tasks are stored in TaskMD format and generate Git Note audit records, supporting multi-agent collaboration and cross-session recovery.

## Agent Orchestration and Integration: OpenSpec Specifications and Automated Workflows

### Agent Orchestration
- pi editor: Execute tasks end-to-end via the /hub-run command, with loop-based queue clearing
- OpenCode: Offers options like @hub-runner (single task) and @hub-orchestrator (batch parallel execution)

### OpenSpec Integration
Convert ideas into OpenSpec specifications via the /opsx-propose command, generate task lists synchronized to the Hub, and enable specification-driven development.

### GitHub Actions Automation
The sync-template.yml and sync-hub.yml workflows automatically synchronize changes, requiring TEMPLATE_REPO_TOKEN and OPENROUTER_API_KEY secrets.

## Quick Start and Toolchain Ecosystem

### Quick Start
1. Clone mcp-task-hub and start the Docker service
2. Clone agent-template as the foundation for new projects
3. Configure openspec/config.yaml and AGENTS.md
4. Enable Git Note functionality

### Toolchain Ecosystem
Integrates open-source tools such as the MCP protocol (open-sourced by Anthropic), TaskMD, OpenSpec, OpenCode, and the pi editor, with links to official documentation provided.

## Limitations and Future Development Directions

### Limitations
- Mode B relies on the OpenRouter API key
- Automated PRs via GitHub Actions require manual review
- Single-machine deployment of the task hub limits team collaboration

### Future Directions
- Support for multi-hub federated architecture
- More fine-grained permission control
- Integration with more AI editors
- Git Note-based intelligent analysis features

## Conclusion: The Paradigm Shift in AI Development Brought by Agent-Kit

Agent-Kit represents a paradigm shift from "AI assistant" to "virtual colleague", providing organizations with a reference for specification-driven, state-externalized, and human-machine collaborative work methods. It serves as a bridge connecting AI potential with engineering practice, driving the future development of autonomous programming.