HarnessKit CLI
A Rust app for macOS, Linux and Windows: skills, MCP servers, plugins, hooks and configs of 17 AI agents in one window, with a security audit
- 478 ★ on the repository
- Last commit
- Apache-2.0
Broad access
Why this level
- The app reads and changes the configs, hooks, MCP and memory of every agent it finds
- Deploying an extension to other agents adds hooks and servers that will run when agents work
- Installation runs a `curl | sh` script, and web mode opens a port on the machine
Install
- Works in
Manual install
curl -fsSL https://raw.githubusercontent.com/RealZST/HarnessKit/main/install.sh | shWeb mode for macOS and Linux. Then run `hk serve` and open the printed address with the token. On Windows use install.ps1 in PowerShell.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The app reads and changes the configs, hooks, MCP and memory of every agent it finds
- Deploying an extension to other agents adds hooks and servers that will run when agents work
- Installation runs a `curl | sh` script, and web mode opens a port on the machine
This is third-party code. Review the repository files before installing.
What it does
HarnessKit brings into one interface what different AI agents keep in different folders and formats. It supports 17 agents, among them Claude Code, Codex, Gemini CLI, Cursor, Copilot, OpenCode, Kiro and Antigravity. For each you see skills, MCP servers, plugins, hooks and agent-first CLIs, plus configs, memory, rules, subagents and ignore files. You can filter all of it by type, agent and source, enable and disable items right from the list and check for updates. Cross-agent deployment shows which agents already have an extension and installs it for the rest with one click, handling format differences itself: JSON, TOML, hook conventions and MCP schemas. Extensions from one repository are grouped into packs for batch operations. A built-in security engine applies 18 static analysis rules and gives each extension a trust score from 0 to 100: safe from 80, low threat from 60 to 79 and needs review below 60. Each finding names the file and line. The app runs as a desktop app for macOS, as a web interface with `hk serve` for any OS and for servers without a screen, and as a console utility.
Who it is for. For developers who use several AI agents and want to manage skills, MCP and configs from one place and audit them for safety.
Good fit when
- A skill or MCP server sits in Claude Code and you want to install it into Codex and Cursor with one click
- You want to see which extensions and hooks have piled up across all agents on the machine and disable the extras
- Before installing someone else's skill you want to run static analysis rules and get a trust score
- Configs and rules files are scattered across global and project folders and you want them in one dashboard
- Agents run on a remote server with no screen and you manage them through the web interface over ssh
Not a fit when
- You have one agent and a few extensions: the agent itself is enough
- You need only a command-line skill installer: the catalog has lighter CLIs for that
- A third-party app must not get access to agent configs holding tokens and keys
- You need final safety guarantees: the trust score is built on static analysis and does not replace reading the code
Example request
Run hk serve and open the list of extensions with a trust score below 60Limitations
The app reads and changes the configs of every agent it finds, so their folders with settings, memory and rules are open to it. Not every extension type is supported for every agent: in the README table some cells carry a dash (for example OpenClaw has no plugins or hooks, and PenguinHarness has only skills). The desktop version is described in the README for macOS and ships as a DMG for Apple Silicon and Intel, while other systems use the web mode. Installation on Linux, macOS and Windows runs through a `curl | sh` or `irm | iex` script. The trust score rests on 18 static analysis rules and is a heuristic. In web mode token authentication is on by default, and the `--no-token` flag turns it off.
How to disable. Close the app or stop `hk serve`. For the desktop version drag HarnessKit from Applications to the trash. The install script places the program itself, so remove the `hk` binary it put on the system.
FAQ
How do I install HarnessKit?
On macOS download the DMG for your chip (Apple Silicon or Intel) from the releases and drag the app to Applications. On macOS, Linux and Windows the web version installs with a script: `curl -fsSL https://raw.githubusercontent.com/RealZST/HarnessKit/main/install.sh | sh`, and for Windows `irm .../install.ps1 | iex`. Then run `hk serve`.
Which agents are supported?
Seventeen: Claude Code, Codex, Gemini CLI, Cursor, Antigravity, Copilot, Devin Desktop, OpenCode, Hermes, Kiro, Oh My Pi, DeepSeek Harness, Grok Build, Qoder CN, OpenClaw, PenguinHarness and Pi. The README table shows which extension types each one gets.
How does the trust score work?
An engine of 18 static analysis rules checks each extension and assigns a score from 0 to 100. Safe from 80, low threat from 60 to 79, needs review below 60. Each finding names a file and line, and for a shared extension the score is computed per agent.
Can I use it on a server with no screen?
Yes. Install HarnessKit on the server with the script, forward the port with `ssh -L 7070:localhost:7070 user@server` and run `hk serve`. Open the address with the token it prints in your local browser. Web mode matches the desktop app in features, except actions such as Open in Finder.
What does HarnessKit do with agent configs and memory?
For each agent it shows configs, memory, rules, subagents and ignore files on a separate page, split by category with scope, path and size. You can expand a file to preview it, and changes to a file show up in the dashboard at once. You can add your own files and folders to tracking.
Discussion
No questions or reviews yet.
Related
CLI toolsTurns a GitHub repo into a package with its own AI agent: an npx command, an MCP server, memory and a signed release
An open-source agent harness in Rust: GUI, CLI, headless and webapp from one binary, with multiple providers, MCP, skills and agent teams
An engineering system for coding agents: 71 agents, 311 skills, hooks, memory, rules and the AgentShield security scanner
The Harness MCP server: one agent starts other Claude Code and Codex sessions on your computer, waits for them, reads their answers and closes them
The /better-harness command reviews how an AI agent's work is set up in a project: goals, checks, delivery and memory, and returns a prioritized fix report
A terminal coding agent built on Codex, tuned for low-cost open models like Kimi K3, DeepSeek and Qwen
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Repository stars | Main difference |
|---|---|---|---|
| HarnessKit (this entry) | no | 478 | |
| SkillKit | no | 1,564 | a CLI skill manager for installing from repositories and translating between agent formats, without managing MCP, hooks and configs |
| OpenSkills | no | 10,811 | a universal loader for SKILL.md skills for any agent, without a graphical interface or audit |
README in short
The README opens with the slogan of one home for every agent and explains the problem: each agent has its own world of extensions, configs, memory and rules in different folders. Sections follow on full management of five extension types with a compatibility table, on configs, memory and rules, the security audit, web mode and an interface with themes. It ends with desktop and web-mode installation, locally and on a server, and a roadmap.