Accelerate Clones, Models, and Package Downloads with Xget

A big clone, a model checkpoint, or a cold npm install can stall your whole session — and the fix differs per platform. Xget is an 8k-star acceleration engine that puts one edge URL in front of code hosting, model and dataset hubs, package registries, container registries, and AI inference providers, adding caching, retries, security headers, and protocol-specific compatibility on the way. Its companion agent skill does the rewiring for you.

Why This Skill Matters

The URL pattern is one substitution: swap the domain and add a platform prefix. GitHub files become https://xget.xi-xu.me/gh/..., Hugging Face becomes https://xget.xi-xu.me/hf/..., PyPI becomes https://xget.xi-xu.me/pypi/.... The README's conversion table documents prefixes for GitHub, GitLab, Gitea, Codeberg, SourceForge, AOSP, Hugging Face, Civitai, arXiv, package registries including npm, PyPI, conda, Maven, NuGet, and Rust crates, Linux distribution mirrors, plus cr for container registries and ip for AI inference providers.

Under the hood the engine adds edge caching with strategy-based durations, up to 3 retries with linear delay, a 30-second request timeout, and a security-header set that includes Strict-Transport-Security and Content-Security-Policy. Git operations skip caching so repository data stays real-time.

The agent skill — installed from the same author's xixu-me/skills repo — turns that table into execution: it prefers running the commands and editing the real config files over printing examples, resolves your Xget base URL in a defined order, and pulls the README's live Use Cases guidance with a helper script.

Installation

npx skills add xixu-me/skills -s xget

The skill lands in your agent's skills directory as xget. Xget itself needs no install if you use the pre-deployed instance xget.xi-xu.me — the README scopes it to evaluation and trial, and says to deploy your own instance for production or availability-sensitive workloads. For turning any supported URL into accelerated form by hand, there is also a web converter at xuc.xi-xu.me.

Real Workflow: Switch npm and Hugging Face to Accelerated Endpoints

1. Install the skill, then state the goal with your base URL:

把这台机器的 npm 依赖下载和 Hugging Face 模型下载切换到 Xget 加速，
基础地址用 https://xget.xi-xu.me，改完各跑一次安装/下载验证。

2. The skill resolves the base URL first — a domain you gave wins, otherwise it reads XGET_BASE_URL from the environment, otherwise it asks whether to set it temporarily or persistently.
3. It classifies the task (package-manager config, AI SDK base URLs, Git, containers) and pulls just the matching README sections via scripts/xget.mjs, its bundled helper for live platform data and Use Cases lookup.
4. It edits the real files — npm registry settings, HF endpoint variables — and runs the verification you asked for instead of stopping at example commands.

The same flow covers Git clones (point clone URLs at the gh prefix) and container pulls, with the README documenting global Git configuration and integrations for mainstream download tools.

Tips

• Set XGET_BASE_URL in your shell profile so the skill skips the base-URL question entirely.
• The skill keeps a https://xget.example.com placeholder for docs and templates — do not treat it as a working endpoint.
• Self-hosting is documented for Cloudflare Workers, Cloudflare Pages, EdgeOne Pages, Vercel, Netlify, Deno Deploy, and Docker/Podman.
• If you self-host, the README says to put the instance behind authentication, IP allowlists, or both unless you explicitly intend to run a public mirror — apply the same logic before routing any private registry traffic through an instance you do not control.

When Not to Use This

The pre-deployed instance is explicitly for evaluation and trial — production or availability-sensitive workloads want your own deployment. And when your upstream is already fast and unmetered, an extra hop buys you nothing; this is a tool for constrained or throttled networks, not a default for every download.

See the leaderboard for more skills.