AI Coding Skills
The techniques, knowledge, and agent skills that make AI-assisted development actually work. Curated for quality, every entry is relevant, documented, and maintained.
How to Use This Hub
Skills are grouped by category. If you're new to AI-assisted development, start with Rapid Prototyping Skills then move to Prompt Engineering. If you're already building, explore Agent Skill Directories and Spec-Driven Development to level up.
Agent Skill Directories
Open directories and registries for installing procedural knowledge into AI coding agents.
Prompt Engineering Skills
The core skill of communicating effectively with AI coding tools, from basic prompting to advanced techniques.
Prompt Engineering for AI Code Tools
The 5 golden rules for talking to AI coding assistants effectively.
Read moreMastering Cursor Composer
Advanced prompting and multi-file editing techniques for Cursor.
Read moreAdvanced Vibe Coding Techniques
Level up from basics, context management, multi-file refactors, and production patterns.
Read moreSpec-Driven Development
Skills for structuring AI work with specifications, plans, and task files, keeping projects on track as they grow.
Rapid Prototyping Skills
Skills for going from idea to deployed prototype as fast as possible using AI builders.
Tracked Skills & Format Specs
Skill packs and open format specs we’ve reviewed and indexed. Click any card for the full write-up.
AGENTS.md
specAGENTS.md is an open format standard for documenting instructions that AI coding agents need to work on software projects. Think of it as a README for agents: a dedicated, predictable place for setup commands, code style, testing instructions, and project conventions. Stewarded by the Agentic AI Foundation under the Linux Foundation, it is used by 60,000+ open-source projects.
DESIGN.md
specDESIGN.md is a format specification from Google Labs for describing a visual identity to coding agents. It pairs machine-readable design tokens (YAML front matter) with human-readable design rationale (Markdown prose), giving agents a persistent, structured understanding of your design system. Ships with an official CLI (@google/design.md) for linting (WCAG contrast, broken refs), diffing, and exporting to Tailwind v3/v4 or DTCG. Still in alpha.
Obsidian Skills
skillOfficial collection of Agent Skills for Obsidian, created by CEO Steph Ango (kepano). Teaches AI agents how to work with Obsidian Markdown (wikilinks, callouts, frontmatter), Bases databases, JSON Canvas whiteboards, the Obsidian CLI, and Defuddle web extraction. Compatible with Claude Code, Codex, and any skills-compatible agent. 13.9k+ GitHub stars.
The State of AI Coding Skills in 2026
What 'AI Coding Skill' Actually Means in 2026
The term moved from informal to concrete in 2025. A skill used to mean any technique that made you better with AI tools. In 2026 it also means a specific artifact: a markdown or YAML file that lives in your repo and teaches a coding agent how to handle a class of problem. Anthropic's Claude Code introduced installable skills mid-2025, Cursor formalized rules (.cursor/rules/*.mdc) around the same time, and directories like Skills.sh consolidated the pattern across tools. The result: you can now check skills into Git, share them with teammates, and ship a workflow alongside the code it produces.
The Three Layers Every Vibe Coder Eventually Builds
Most developers who stick with AI tools end up with the same skill stack regardless of which tool they started with. Layer one is prompt patterns (the 5 to 10 reusable openers you type every day, codified). Layer two is repo-specific rules (the conventions your codebase uses, encoded in .cursor/rules/ or skills.md so the agent stops re-asking). Layer three is workflow skills (spec-first, test-after, debug-from-trace), the higher-order processes that determine whether the agent ships working code or rabbit-holes for an hour. The categories on this page map roughly to these layers.
Why Spec-Driven Beats Vibes for Anything Real
The fastest way to get an AI agent to ship the wrong thing is to describe what you want in one sentence and hit enter. The fastest way to get it to ship the right thing is to write a one-page spec first. GitHub Spec Kit and similar tools formalize this: spec, then plan, then tasks, then implementation. The investment is 10 to 20 minutes upfront for hours of saved rework. The spec also becomes shareable, which matters the moment more than one person works on the same agent-built feature.
Skills vs Plugins vs Tools
Three adjacent concepts and most people conflate them. A tool is the software (Cursor, Claude Code, Bolt). A plugin extends a tool (an MCP server, a VS Code extension, a Cursor extension). A skill is portable knowledge (a markdown file, a prompt pattern, a process). Tools and plugins live in the tool's settings; skills live in your repo and travel with the codebase. If you want to share a way of working with your team, ship a skill. If you want everyone using the same software, ship a tool recommendation. The /plugins and /tools pages cover the other two.
Frequently Asked Questions
What are AI coding skills?
AI coding skills are specific techniques, knowledge packages, or procedural instructions that make AI coding agents more effective. They range from prompt engineering patterns to installable skill files (like Skills.sh) that give agents domain-specific expertise.
How are skills different from tools?
Tools are the software products you use (Cursor, Claude Code, Bolt.new). Skills are the techniques and knowledge that make you, and your AI agents, more effective with those tools. Think of tools as instruments and skills as the ability to play them well.
How does curation work on this page?
Every skill listed here meets our inclusion criteria: it must be directly relevant to building software, have a clear primary use-case, have accessible documentation, and show evidence of active maintenance. We exclude abandoned, vague, or low-trust entries.
What is a Claude skill or Cursor skill?
Installable skill files are small markdown or YAML packages that teach an AI coding agent a specific technique or domain. Claude Code skills (via Skills.sh and similar directories) and Cursor rules (.cursor/rules/*.mdc) are the two most common formats. They live in your repo, version-control like code, and apply to every session.
Do skills replace prompt engineering?
No, they encode it. A skill is reusable prompt-engineering: instead of typing the same setup instructions every session, you save them once and the agent reads them automatically. Hand-written prompts still matter for one-off tasks; skills matter when a pattern repeats.
Where should I start if I'm new to vibe coding?
Read the foundational guides in the Rapid Prototyping section first (Build Your First App, How to Vibe Code). Then move to Prompt Engineering. Skill directories and spec-driven workflows are more valuable once you have a working baseline and want to scale up.
Are these skills tool-specific?
Mixed. Some skills are tied to specific tools (Cursor Composer techniques, Claude Code skill packs). Others (prompt engineering, spec-driven development) apply to every AI coding agent. The category pages on each card tell you which.