One‑Click Sync of AI Coding Skills Across 15 Platforms with PromptHub

Background and Motivation

Developers often keep SKILL.md files scattered across Claude Code, Cursor, Codex and other AI coding tools, requiring manual synchronization that is error‑prone and time‑consuming. PromptHub addresses this pain point by providing a unified desktop client that runs on Windows, macOS and Linux.

Core Features

Prompt Management

Folders, tags, and drag‑and‑drop sorting give full CRUD capabilities. Template variables such as {{框架}} or {{语言}} can be defined, prompting a fill‑in dialog during testing. Full‑text search uses SQLite FTS5, and markdown rendering includes syntax highlighting, inline image and attachment preview. Double‑clicking a card opens the System Prompt and User Prompt for quick edits.

Quick Add

Three entry points let users add prompts manually, analyze existing prompts, or generate them with AI. AI‑generated prompts are of average quality but still better than starting from a blank input.

Skill Store

The store ships with over 20 curated skills from Anthropic and OpenAI and allows users to add sources from GitHub repositories, the skills.sh community, or local directories. Installing a skill automatically inserts the corresponding SKILL.md into each tool’s expected location. The store supports fifteen to sixteen platforms, including Claude Code, Cursor, Windsurf, Codex, Kiro, Kilo Code, Gemini CLI, Cline, Qoder, CodeBuddy, Trae, Trae CN, OpenCode, and others.

Installation Options

During installation users can choose between a symlink (shared source file across platforms) or a copy (isolated files). Symlinks propagate changes instantly; copies keep platforms independent.

Local Scanning and AI Assistance

PromptHub scans the local filesystem for existing SKILL.md files, presents previews, and enables batch import. An AI translator can generate side‑by‑side translations of a skill file, and an AI reviewer checks content before import. GitHub tokens are used to avoid anonymous‑rate‑limit blocks.

Rules Management

Rules such as CLAUDE.md or .cursor/rules are consolidated into a single panel with search, edit, and version‑snapshot capabilities. Snapshots can be previewed, restored, exported as ZIP, synchronized via WebDAV, self‑hosted sync, or imported/exported through the web.

Project and Agent Assets

Each project gets an independent skill workspace that automatically scans common directories ( .claude/skills, .agents/skills, skills, .gemini) and allows preview‑and‑import of discovered assets.

AI Testing and Multi‑Model Comparison

PromptHub integrates OpenAI, Anthropic, Gemini, Azure, and custom endpoints. The same prompt can be sent in parallel to multiple models, with results displayed side by side. Version 0.5.8 introduces a three‑column vendor‑priority configuration separating vendor instance, model capability, and business routing. Image‑prompt reverse engineering is also supported, producing structured generation prompts from an input image.

Version Control

All prompts are automatically versioned with history, diff highlighting, and one‑click rollback, similar to Git but lighter. Skills support named versions, and Rules snapshots can be previewed and restored. Content hashes detect changes in remote SKILL.md files, triggering conflict warnings. Community skill updates are delivered via GitHub Releases.

Sync, Backup, and Security

Data is stored locally first; a full backup is created as .phub.gz. Synchronization uses WebDAV and works with Nutstore, Nextcloud, or a self‑deployed web server. Automatic pull and scheduled push run in the background, with only one active sync source to avoid multi‑source conflicts. The application is protected by a main password encrypted with AES‑256‑GCM; a beta encrypted private folder works offline. Seven UI languages and customizable animations are available.

CLI and Self‑Hosted Web

The CLI, intended for scripting, covers six categories (prompt, folder, rules, skill, ai, workspace) with --help support; it is built from source only. The web version runs in Docker, suitable for NAS, VPS, or LAN, and uses a .env file to set JWT_SECRET, data directory, and port 3871. The first user becomes the administrator. An experimental Cloudflare Workers build runs the API on Workers, stores data in D1, and media in R2.

Installation

Pre‑built installers for Windows, macOS, and Linux are available on GitHub Releases. macOS users can also install via Homebrew:

brew tap legeling/tap
brew install --cask prompthub

Because the app lacks notarization, macOS may show “cannot be verified”; running

sudo xattr -rd com.apple.quarantine /Applications/PromptHub.app

removes the warning.

Evaluation

If you only use one or two AI coding tools and have few prompts, PromptHub may add unnecessary complexity. However, for users juggling multiple tools (e.g., Claude Code, Cursor, Codex) with dozens of prompts and skills, it eliminates repetitive manual syncing and keeps data under personal control. Compared with Cherry Studio (dialogue‑focused) and ChatWise (chat‑oriented), PromptHub’s sole purpose is to manage fragmented AI‑coding assets across platforms.

Source

https://github.com/legeling/PromptHub