Harness AI Kit

harness-ai-kit

A package manager and composition layer for AI agent assets.

harness-ai-kit installs, resolves, locks, validates, and synchronizes Skills,
CLIs, MCPs, plugins, hooks, subagents, and loops across Codex, Claude Code,
Cursor, Kiro, and DeepSeek Harness (dsh).

Quickstart · Asset Catalog · Usage Scenarios · Concepts · Roadmap · Changelog · 中文文档

Why

AI teams collect useful prompts, Skills, CLIs, and MCP servers quickly. The
hard part is making that collection reproducible: knowing what is installed,
which versions work together, where it is materialized, and how a teammate can
get the same working state without copying runtime directories.

harness-ai-kit makes the project manifest the source of truth:

The lock records resolved sources and checksums. Installation uses a staging
directory and only replaces the runtime target after the selected assets are
ready. The result is useful for one developer, and essential when a team needs
the same AI engineering environment across projects and runtimes.

The REMIX Method

This project is a composition layer, not another vertical AI toolkit. When a
focused community Skill, CLI, or MCP already solves a problem, the preferred
path is to compose it, pin it, and make it reproducible instead of rebuilding
it.

- Reuse proven assets from the community or your own repositories.
- Extend an asset when a small, explicit adaptation is enough.
- Mix Skills, CLIs, MCPs, and runtime assets into one workflow.
- Integrate the selected assets through a typed manifest and lockfile.
- eXecute the same declared environment across supported runtimes.

No Lock-In

An installed Skill is still a normal directory of Markdown and metadata. You
can read it, copy it, or install it directly from its Git repository without
using this tool.…