扩展与可改造设计

OpenClaw is built to be hackable, meaning you can open it up and change how it works, not just use it as-is. 🛠️

You add new integrations through an extensions directory, the place where extra capabilities live alongside the core.

Each folder under extensions/* can hold its own integration, so you grow OpenClaw by adding pieces rather than rewriting it.

The source ships as a single monorepo, a pnpm workspace that keeps all the parts of OpenClaw together in one place.

That workspace is managed with pnpm, so the packages and extensions link together cleanly during development.

Because it is open, you can modify prompts the assistant uses, tuning how it thinks and responds to your own taste.

The whole project is open-source under the MIT license, so reading and changing the code is fully within your rights. 📖

OpenClaw integrates with coding-agent runtimes in the Claude Code or Codex style, fitting into how developers already work.

Skills add abilities to the assistant, while extensions add integrations to the system itself. Together they expand what OpenClaw can do.

Every extension you add keeps the local-first promise, running on your own hardware under your own rules. 🔒

Between extensions, prompt edits, and skills, you can shape OpenClaw into an assistant that fits exactly how you want to work. 🌟

Let us check how you extend OpenClaw.

You learned OpenClaw is a hackable MIT monorepo. Add integrations via extensions/*, modify prompts, and pair it with coding runtimes. 🚀