🪨 why use many token when few token do trick — Claude Code skill that cuts 65% of tokens by talking like caveman
AI agent setup manager for CLI, desktop, and web
HarnessKit keeps agent-specific skills, MCP servers, plugins, hooks, CLIs, configs, memory, and rules in one place. It works directly in each agent’s native directories, so you can manage, audit, enable, disable, and deploy setups without copying them into a separate folder.
Builders who use multiple AI coding agents and want one place to manage their files and extensions.
You can keep every agent’s setup organized, audit it, and deploy it across projects without redoing the same work by hand.
What it does
Unified extension management
Lists and manages skills, MCP servers, plugins, hooks, and agent-first CLIs across supported agents.
Config and memory tracking
Finds global and project config files, memory, rules, subagents, and ignore files, then shows them in per-agent dashboards.
Security audit and trust scores
Scans extensions with static rules, shows a trust score, and surfaces permissions, findings, and file paths.
Marketplace browsing
Browses skills from skills.sh, MCP servers from Smithery, and agent-first CLI tools, with install and preview support.
Kit packaging
Bundles skills, MCP servers, rules, and memory files into portable `.hk-kit.zip` packages for reuse across projects.
CLI and web mode
Uses the `hk` command for terminal work and `hk serve` to open the same app in a browser.
How to get it
- 1Install HarnessKit
# macOS / Linux curl -fsSL https://raw.githubusercontent.com/RealZST/HarnessKit/main/install.sh | sh
- 2Run
# Windows (PowerShell) irm https://raw.githubusercontent.com/RealZST/HarnessKit/main/install.ps1 | iex
- 3Start the web interface
hk serve
- 4Install HarnessKit on the server
# macOS / Linux server ssh user@your-server curl -fsSL https://raw.githubusercontent.com/RealZST/HarnessKit/main/install.sh | sh exit
- 5Run
# Windows server ssh user@your-server irm https://raw.githubusercontent.com/RealZST/HarnessKit/main/install.ps1 | iex exit
- 6Start the web interface
ssh -L 7070:localhost:7070 user@your-server hk serve
README
HarnessKit
English · 简体中文
One home for every agent.
A free, open-source app to manage all your AI coding agents — desktop, CLI, or web.
Why • Features • Get Started • Roadmap
Every supported agent shows on the Overview by default, installed or not. In Settings → Agent Paths, switch to "Detected only" to hide and disable undetected agents, or flip a single agent's "Enabled" toggle off.
Why HarnessKit?
Every agent, a different world. Extensions, configs, memory, and rules — scattered across different directories, in different formats, with different conventions.
HarnessKit brings them all under one roof — see, secure, and manage everything across every agent, from one place.
Key Features
🧩 Full Suite Extension Management
HarnessKit manages all five extension types from a unified interface — Skills, MCP Servers, Plugins, Hooks, and Agent-first CLIs.
| Agent | Skills | MCP | Plugins | Hooks | Agent-first CLIs |
|---|---|---|---|---|---|
| Claude Code | ✓ | ✓ | ✓ | ✓ | ✓ |
| Codex | ✓ | ✓ | ✓ | ✓ | ✓ |
| Gemini CLI | ✓ | ✓ | ✓ | ✓ | ✓ |
| Cursor | ✓ | ✓ | ✓ | ✓ | ✓ |
| Antigravity | ✓ | ✓ | — | — | ✓ |
| Copilot | ✓ | ✓ | ✓ | ✓ | ✓ |
| Devin Desktop | ✓ | ✓ | — | ✓ | ✓ |
| OpenCode | ✓ | ✓ | ✓ | — | ✓ |
| Hermes | ✓ | ✓ | ✓ | ✓ | ✓ |
| Kiro | ✓ | ✓ | — | ✓ | ✓ |
| Oh My Pi | ✓ | ✓ | ✓ | — | ✓ |
| DeepSeek Harness | ✓ | ✓ | ✓ | — | ✓ |
| Grok Build | ✓ | ✓ | ✓ | ✓ | ✓ |
* "—" indicates the agent currently does not support this extension type. Devin Desktop support keeps legacy Windsurf paths compatible.
- Smart organization — Filter by type, agent, or source, and search by name. Extensions from the same repo are automatically grouped into packs for batch management.
- Full visibility — Every extension shows its agents, permissions, trust score, and status at a glance. Open the detail panel for per-agent file paths, directory structure, and audit findings.
- Effortless management — Enable or disable right from the list. Check for updates across all extensions with one click.
- Cross-agent deployment — See which agents have the extension and which don't — deploy to any missing agent with one click. HarnessKit handles the format differences between agents (JSON, TOML, hook conventions, MCP schemas) automatically.
🤖 Agent Configs, Memory & Rules
HarnessKit manages every agent's Configs, Memory, Rules, Subagents, and Ignore files from one place. Currently supporting 13 agents: Claude Code, Codex, Gemini CLI, Cursor, Antigravity, Copilot, Devin Desktop, OpenCode, Hermes, Kiro, Oh My Pi, DeepSeek Harness, and Grok Build.
- Config file tracking — Automatically discovers every agent's config files — both global and per-project. Add your project directories or custom paths and HarnessKit picks them up alongside the global ones.
- Per-agent dashboard — Each agent gets its own page with all files organized by category, showing scope, path, file size, and a summary of installed extensions. Expand any file to preview its content right in the app.
- Custom paths — Add any file or folder to an agent's dashboard for tracking. Useful for custom configs or scripts that HarnessKit doesn't auto-discover — they show up alongside everything else with the same live preview.
- Real-time detection — The moment a config file is modified, the dashboard reflects it.
🛡️ Security Audit & Permission Transparency
Every extension is scanned by a built-in security engine with 18 static analysis rules and receives a Trust Score (0–100), grouped into three tiers — Safe (80+), Low Risk (60–79), and Needs Review (below 60). A dedicated Audit page lets you search, filter by tier, and drill into every finding.
- One-click audit — Run a full security scan across all extensions with a single click. The dashboard shows how many extensions were scanned and when the last audit ran.
- Precise tracing — Every finding pinpoints the exact file and line number, so you can trace the issue immediately.
- Per-agent scanning — Even if multiple agents share the same extension, each agent's copy is audited independently — because versions can drift, and a safe copy on one agent doesn't guarantee safety on another.
- Permission transparency — Every extension's permissions are surfaced across five dimensions — filesystem paths, network domains, shell commands, database engines, and environment variables. You see exactly what each extension can reach before you decide to keep it.
🏪 Marketplace Ecosystem
Discover, evaluate, and install — three marketplaces in one, each with trending lists and search:
- Skills — Browse and install from the skills.sh registry. Also supports install from Git URL or local directory.
- MCP Servers — Browse the Smithery registry of Model Context Protocol servers.
- Agent-first CLI — Discover CLI tools built specifically for agents — the newest frontier of the agent extension ecosystem.
Every listing shows its description, install count, and source. For skills, you can preview the documentation, check third-party security audit scores before installing, and install to any agent with one click — HarnessKit tracks the source so you always know where each extension came from.
🔀 Project-Level Management
The sidebar scope picker switches between Global, All scopes, or any registered project. Agents, Extensions, and Audit all filter by the active scope — per-project setups are managed independently of your global config.
📦 Kits
Pack a curated set of skills, MCP servers, rules and memory files into a portable Kit — then deploy the whole bundle to any project with one click. Skip the setup churn every time you spin up a new project.
- Compose once, reuse everywhere — Build a Kit from your existing extensions, rules, and memory files. Pick a target agent at install time and HarnessKit writes everything to the right places.
- Multi-project ready — Install the same Kit to as many projects as you need. The detail drawer shows where each one is currently deployed, and removing it from a project cleans up cleanly.
- Portable bundles — Export any Kit as a self-contained
.hk-kit.zipto share with teammates or carry across machines. Import is one click. - Origin tracking — Kit-installed extensions merge with their marketplace origin in the Extensions list, so you always know where each extension came from.
📂 In-Place Management
HarnessKit works directly with your agents' native directories instead of copying them into a managed folder — no shadow copies, no sync conflicts.
- Native directories — Reads and writes directly to each agent's own config directory. Your files stay exactly where they are.
- Non-destructive operations — Enabling or disabling an extension is a simple file rename in place. Nothing is moved or duplicated.
- Zero lock-in — Uninstall HarnessKit and everything is exactly where it was. No migration, no cleanup needed.
⌨️ CLI Support
HarnessKit ships a standalone command-line interface (hk) for terminal-first workflows, available on macOS, Linux, and Windows:
$ hk status
Agents 13 detected (claude · codex · gemini · cursor · antigravity · copilot · windsurf · opencode · hermes · kiro · omp · dsh · grok)
Extensions 136 total (124 skills · 2 mcp · 8 plugins · 1 hooks · 1 clis)
$ hk list --kind skill --agent claude # filter by type and agent
$ hk list --kind mcp --agent grok # Grok Build MCP servers
$ hk audit # security audit with trust scores
$ hk enable my-skill # enable by name
$ hk disable --pack owner/repo # batch disable by source
🌐 Web Mode
The same full-featured UI that runs in the desktop app is also available as a web interface — served directly from the hk CLI binary. No extra dependencies, no separate install.
$ hk serve
HarnessKit Web UI [my-host] running at http://127.0.0.1:7070/?token=a1b2c3…
This makes HarnessKit usable on Linux servers, HPC clusters, or any headless machine where a desktop app isn't an option. Web mode has full feature parity with the desktop app — the only difference is that file-system operations (like "Open in Finder") are desktop-only. See Getting Started for setup instructions.
✨ Thoughtful & Interactive UX
- 💡 Tip of the Day — The Overview dashboard surfaces contextual tips for each detected agent from a community-maintained library. Learn shortcuts and best practices as you work.
- 📊 Dynamic Activity Feed — Agent Activity and Recently Installed timelines capture every config change, extension install, and agent event in real time.
- ⚡ Quick Actions — One-click View Agents, Run Audit, Check Updates, and Marketplace access right from the dashboard.
- 🎯 Playful Touches — Smooth animations and micro-interactions throughout the app make daily use feel alive.
- 🎨 Themes — Multiple themes with Light, Dark, and System mode support.
Getting Started
Requirements: At least one supported AI coding agent installed.
🖥️ Desktop App (macOS)
-
Download the DMG for your architecture from the latest release:
Chip File Apple Silicon (M1/M2/M3/M4) HarnessKit_x.x.x_aarch64.dmgIntel HarnessKit_x.x.x_x64.dmg -
Open the DMG and drag HarnessKit to the Applications folder.
-
Launch HarnessKit. It will automatically detect your installed agents and scan their extensions.
Already installed? Open Settings → Check for Updates to upgrade in-app.
🌐 Web Mode (macOS / Linux / Windows)
Local machine
-
Install HarnessKit:
# macOS / Linux curl -fsSL https://raw.githubusercontent.com/RealZST/HarnessKit/main/install.sh | sh# Windows (PowerShell) irm https://raw.githubusercontent.com/RealZST/HarnessKit/main/install.ps1 | iex -
Start the web interface:
hk serveThen open the
http://localhost:7070/?token=…URL thathk serveprints. Auth is on by default; the token is saved, so next timehttp://localhost:7070just works. On a trusted single-user machine,hk serve --no-tokenskips the token entirely.
Remote server
-
Install HarnessKit on the server:
# macOS / Linux server ssh user@your-server curl -fsSL https://raw.githubusercontent.com/RealZST/HarnessKit/main/install.sh | sh exit# Windows server ssh user@your-server irm https://raw.githubusercontent.com/RealZST/HarnessKit/main/install.ps1 | iex exit -
Start the web interface:
ssh -L 7070:localhost:7070 user@your-server hk serveThen open the
http://localhost:7070/?token=…URL thathk serveprints, in your local browser. Auth is on by default; the token is saved, so next timehttp://localhost:7070just works. Keep the SSH session running while you use HarnessKit.
Tip: Managing several remote nodes? Start each with
hk serve --name <label>(e.g.--name my-macbook). The label shows in the sidebar and the browser tab title, so multiple tabs are easy to tell apart. Defaults to the machine hostname.
Manual download — if you prefer not to use the install script, or your machine doesn't have curl
Download the binary for your platform from the latest release (referred to as <file> below):
| Platform | File |
|---|---|
| macOS (Apple Silicon) | hk-macos-arm64 |
| macOS (Intel) | hk-macos-x64 |
| Linux (x64) | hk-linux-x64 |
| Linux (ARM64) | hk-linux-arm64 |
| Windows | hk-windows-x64.exe |
Local machine:
-
Install HarnessKit:
# macOS / Linux chmod +x <file> mkdir -p ~/.local/bin mv <file> ~/.local/bin/hk# Windows (PowerShell) New-Item -ItemType Directory -Force -Path "$env:USERPROFILE\.local\bin" | Out-Null Move-Item <file> "$env:USERPROFILE\.local\bin\hk.exe" -
Start the web interface:
hk serveThen open the
http://localhost:7070/?token=…URL thathk serveprints. Auth is on by default; the token is saved, so next timehttp://localhost:7070just works. On a trusted single-user machine,hk serve --no-tokenskips the token entirely.
Remote server:
-
Upload and install the binary on the server:
scp <file> user@your-server:~/ ssh user@your-server chmod +x ~/<file> mkdir -p ~/.local/bin mv ~/<file> ~/.local/bin/hk exit -
Start the web interface:
ssh -L 7070:localhost:7070 user@your-server hk serveThen open the
http://localhost:7070/?token=…URL thathk serveprints, in your local browser. Auth is on by default; the token is saved, so next timehttp://localhost:7070just works. Keep the SSH session running while you use HarnessKit.
Updating
Re-run the install script from Local machine or Remote server above — both scripts overwrite the existing hk binary. After updating, restart hk serve to pick up the new version.
If you installed via manual download, grab the latest binary from releases and replace your existing hk (or hk.exe).
⌨️ CLI (macOS / Linux / Windows)
If you've already installed HarnessKit via the Web Mode steps above, the CLI is ready to use — it's the same hk binary.
See CLI Support above for the full list of commands.
Roadmap
- 🤖 More Agents — OpenClaw and more
- ⌨️ CLI Enhancements — More commands and richer functionality for
hk
Contributing
Contributions are welcome! See CONTRIBUTING.md for local setup, project structure, and PR guidelines.
License
This project is licensed under Apache-2.0.
Artwork (public/icons/ and src/components/shared/agent-mascot/) is All Rights Reserved and is not covered by the Apache-2.0 license.
All product names, logos, and trademarks are property of their respective owners. HarnessKit is an independent project, not affiliated with or endorsed by any agent vendor.
Files in the repo
- .github
- crates
- media
- public
- scripts
- src
- .gitignore
- biome.json
- build.sh
- Cargo.lock
- Cargo.toml
- CONTRIBUTING.md
- index.html
- install.ps1
- install.sh
- LICENSE
- package-lock.json
- package.json
- README.md
- README.zh-CN.md
- tsconfig.json
- vite.config.ts
- vitest.config.ts
Discussion (0)
Ask about usage, or say what you built with itSign in to join the discussion.
No comments yet. Be the first to say what this is good for.
More tools
The best-benchmarked open-source AI memory system. And it's free.
Orca is the ADE for working with a fleet of parallel agents. Run any coding agent with your own subscription. Available on desktop, mobile and remote runtime.

A cross-platform desktop All-in-One assistant for Claude Code, Codex, OpenCode, OpenClaw, Grok Build & Hermes Agent. Only official website: ccswitch.io
Never stop coding. Free MIT AI gateway: one endpoint, 352 providers (150+ free), 1200+ models Kimi, Claude, GPT, Gemini, GLM, DeepSeek, MiniMax. Works with Claude Code, Codex, Cursor, OpenCode, Cline & Copilot. Quota-aware auto-fallback, RTK+Caveman compression saves 15-95% tokens, MCP/A2A, Desktop/PWA. Built by 550+ contributors
Compress tool outputs, logs, files, and RAG chunks before they reach the LLM. 20% fewer tokens for coding agents, 60-95% fewer tokens for JSON, same answers. Library, proxy, MCP server.