← Back to Skills Marketplace
119
Downloads
0
Stars
0
Active Installs
1
Versions
Install in OpenClaw
/install universal-primitives
Description
通用执行器架构原理：大语言模型只需文件增删改查和执行脚本两个基元工具，配合其天然的文字理解能力，就能从"无所不知只能输出文字"跃迁到"无所不能能控制任何软件硬件"。当用户讨论AI能力边界、工具设计哲学、Agent架构原理、LLM与操作系统交互、如何让AI控制软件硬件等主题时触发。
Usage Guidance
This skill is a conceptual how‑to for giving an LLM the ability to read/write files and execute commands; the README itself is not malicious but it contains actionable recipes that can lead to running arbitrary code, installing software, or exposing credentials. Before installing or invoking: (1) don't let the agent autonomously run scripts or install services on your machine; (2) never paste secrets or credentials the agent asks for without understanding why and where they will be used; (3) if you test code from the agent, run it only in a tightly sandboxed, ephemeral environment you control (ephemeral VMs, isolated containers, or trusted online sandboxes) and review the code first; (4) avoid allowing the agent to create persistent background services or modify system startup; (5) if you want only conceptual discussion, restrict the agent so it cannot produce executable artifacts or ask users to perform system operations. These precautions reduce the chance that the skill's operational advice is used to perform risky actions.
Capability Analysis
Type: OpenClaw Skill
Name: universal-primitives
Version: 1.0.0
The skill `universal-primitives` (in `SKILL.md`) provides a conceptual framework for AI agents to achieve 'universal' control through file manipulation and command execution. It includes 'Environment Bootstrapping' strategies that explicitly instruct the agent on how to circumvent environment restrictions to gain execution capabilities (e.g., using online sandboxes or guiding users to run code). While no direct malicious payload is present, the instructions to bypass security boundaries and establish execution channels represent high-risk logic for an autonomous agent.
Capability Assessment
Purpose & Capability
The name and description match the SKILL.md content: this is a design/architecture primer about using two primitives (file ops + command execution) and how to bootstrap them. No requested binaries, env vars, or installs are inconsistent with that purpose.
Instruction Scope
The instructions explicitly tell an agent how to obtain or simulate file and execution primitives by (a) instructing users to copy/paste and run scripts, (b) using online code-execution sandboxes (Replit/Colab/etc.), (c) calling REST APIs or browser automation to create/execute code, and (d) using code interpreters to simulate a filesystem and exec. These are broad, actionable steps that can lead the agent to request or cause arbitrary code execution, software installation, or data transmission to external endpoints. The SKILL.md gives the agent wide discretion to create persistent execution channels and to move beyond pure explanation into operational commands — a scope that goes beyond passive documentation.
Install Mechanism
This is an instruction-only skill with no install spec and no code files. Nothing is written to disk by the skill itself, which minimizes direct install risk.
Credentials
The skill declares no required credentials or config paths. However the runtime guidance prescribes interacting with external services (HTTP APIs, online sandboxes, browser drivers) and instructing end users to run commands — all of which commonly require credentials or tokens in practice. The skill does not declare or constrain how such credentials should be handled, so an agent following these instructions could ask for or attempt to use secrets not declared here.
Persistence & Privilege
Although the skill itself does not request persistent privileges or set always:true, its recommended self-bootstrap and recursive 'write-and-run' patterns encourage creation of scripts/services that could be installed to run unattended or persist across sessions. That capability is not represented in the metadata and could enable long-lived execution channels if an agent (or a user following its instructions) implements them.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install universal-primitives - After installation, invoke the skill by name or use
/universal-primitives - Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.0
- Initial release of universal-primitives skill.
- Explains the principle that LLMs need only file operations and command execution ("two primitives") to become general-purpose executors.
- Details how these two primitives empower LLMs to create, modify, and run any software or control hardware indirectly.
- Describes the minimal toolset philosophy and how Agent architectures reduce to combinations of these primitives.
- Provides strategies for "bootstrapping" these primitives even in restricted Agent or user environments.
Metadata
Frequently Asked Questions
What is Universal Primitives?
通用执行器架构原理：大语言模型只需文件增删改查和执行脚本两个基元工具，配合其天然的文字理解能力，就能从"无所不知只能输出文字"跃迁到"无所不能能控制任何软件硬件"。当用户讨论AI能力边界、工具设计哲学、Agent架构原理、LLM与操作系统交互、如何让AI控制软件硬件等主题时触发。 It is an AI Agent Skill for Claude Code / OpenClaw, with 119 downloads so far.
How do I install Universal Primitives?
Run "/install universal-primitives" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Universal Primitives free?
Yes, Universal Primitives is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does Universal Primitives support?
Universal Primitives is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Universal Primitives?
It is built and maintained by 波动几何 (@wangjiaocheng); the current version is v1.0.0.
More Skills