God-Skills: A Curated Personal Repository for Building LLM 'God-Level Skills'

The god-skills repository on GitHub is a practice-oriented curated collection of Agent Skills, featuring high-quality prompt engineering validated in real-world work. It follows three core principles: practice first, reusability, and clear provenance. Adhering to the pragmatism philosophy of 'ready to use out of the box', it helps developers enable LLMs to efficiently complete complex tasks.

In modern AI programming environments like Claude Code, Cursor, and Codex, a Skill is a set of discoverable and callable structured instructions and resources (usually accessed via a SKILL.md file). It guides the model to perform specific tasks according to an agreed-upon process, reducing deviations and rework caused by ad-hoc operations. Simply put, Skills are like 'professional toolkits' for AI—you don’t need to describe requirements from scratch every time; just load the validated workflow directly.

Unlike comprehensive repositories, the god-skills repository proposes three core principles: 1. Practice First: Only include skills validated in real projects; reject theoretical solutions. 2. Reusability: Clear processes that can be triggered repeatedly, not one-off conversation tricks. 3. Clear Provenance: Third-party skills retain licenses and sources; self-written skills are fully open. This 'shortlist' strategy ensures every entry is a practical skill the author has trusted for a long time.

A typical skill currently publicly included in the repository is the 'Multi-Channel Promotion Skill for GitHub Open-Source Projects'. It addresses the pain point of efficient promotion after an open-source project is released, providing an end-to-end process: automatically generating promotion copy adapted to various platforms, completing multi-site posting with a logged-in browser, and guiding README screenshots and release key points for each site—embodying the pragmatism of 'ready to use out of the box'.

Different AI clients load Skills in different ways: Claude Code uses /plugin or the plugin market for installation, or the /plugin add command with a local path; Cursor places Skills in the project or globally agreed skills directory; Codex/other CLIs usually place them in user-level directories like $CODEX_HOME/skills. General steps: Clone the repository → Copy the required Skill directory → Place it in the specified location of the target environment → Enable it.

The author specially reminds that some Skills may contain scripts or guide the model to execute shell commands. Before use, you need to: carefully read the instructions and license agreement, independently assess whether it is suitable for the current environment, and ensure compliance with organizational requirements.

god-skills represents a paradigm shift—from writing ad-hoc prompts to accumulating reusable skill assets. For developers who want to systematically improve the efficiency of using AI tools, the idea of 'curated rather than massive' is worth learning. True 'god-level' skills are not about quantity, but about those reliable tools that are 'ready to use out of the box'.