RealZST/HarnessKit

More than a skill manager — manage skills, MCP servers, plugins, hooks, CLIs, configs, memory & rules across every AI coding agent. 🌟 Star if you like it!

HarnessKit – One Home for Every AI Coding Agent

What it is – HarnessKit is a free, open‑source desktop (macOS, Linux, Windows) and web application that lets you discover, configure, audit, and manage the extensions, settings, memory files and rules of many AI‑coding agents (Claude Code, Codex, Gemini CLI, Cursor, Copilot, etc.) from a single place.

Why it matters – Each AI coding agent stores its own extensions, config files and runtime data in different directories and formats. HarnessKit unifies these scattered resources, giving you a clear overview, bulk actions, and security checks without moving files around.

Core capabilities

Area	What HarnessKit does
Extension management	Handles five extension types – Skills, MCP servers, Plugins, Hooks, and Agent‑first CLIs – for 15+ supported agents. You can enable/disable, update, and deploy extensions across agents, with automatic format conversion (JSON, TOML, etc.).
Config/Memory/Rules	Detects global and per‑project config, memory and rule files for each agent. Shows them in a per‑agent dashboard with live previews and lets you add custom paths.
Security audit	Built‑in static‑analysis engine (18 rules) gives every extension a Trust Score (0‑100) and categorises it as Safe, Low‑Risk or Needs Review. Findings point to exact file/line locations and list permissions (filesystem, network, shell, DB, env).
Marketplace	Integrated browsers for three registries – skills.sh (skills), Smithery.ai (MCP servers), and a catalog of Agent‑first CLIs. Install with one click; source tracking is retained.
Project scopes	Switch between a global view or any registered project, keeping per‑project setups isolated.
Kits	Bundle a set of skills, MCP servers, rules and memory files into a portable `.hk-kit.zip`. Importing a Kit deploys everything to the chosen agent(s) in one step.
In‑place operation	Reads/writes directly in each agent’s native directories; enabling/disabling is just a file rename, so uninstalling HarnessKit leaves nothing behind.
CLI (`hk`)	Provides terminal commands for status, listing, auditing, enabling/disabling extensions, etc., on all platforms.
Web mode	The same UI can be served from the `hk serve` command, enabling use on headless servers or clusters.
UX touches	Tips of the day, activity feed, quick actions, themes (light/dark/system) and smooth animations.

Getting started (quick)

Desktop – Download the appropriate DMG or installer from the latest release, drag to Applications, launch. The app auto‑detects installed agents.
Web / CLI – Run the install script (curl … | sh or PowerShell equivalent) to place the hk binary in ~/.local/bin. Then start the UI with hk serve and open the printed URL, or use commands like hk status, hk list --kind skill --agent claude, hk audit.

Who should use it?

Developers who work with multiple AI coding assistants and want a single pane of glass for extensions and configs.
Teams that need to audit the security of agent extensions across many machines.
Anyone who wants to package a reproducible set of agent extensions (Kits) for onboarding new projects.

License & community

Licensed under Apache‑2.0 (code). Artwork is proprietary.
Contributions are welcomed via the usual pull‑request workflow.

Bottom line: HarnessKit is a practical management layer for the growing ecosystem of AI coding agents, turning a chaotic set of files and plugins into an organized, auditable, and share‑able workflow.