Agent skills as files: a 1,503-star recipe box for AI workflows

What it does

This repository catalogs reusable instruction sets for AI agents. Each skill is a self-contained folder under skills/ that opens with a SKILL.md file defining when to invoke the skill and what workflow to follow. The collection covers agent orchestration, web research, structured thinking, documentation, and machine setup.

The interesting bit

Instead of treating prompts as disposable chat context, the project packages them as version-controlled infrastructure. By standardizing each skill into a folder with a manifest, it nudges agent behavior closer to a callable library than a lucky conversation.

Key highlights

• Categories include agent orchestration, skill authoring, research, documentation, and ops
• Every skill carries a SKILL.md manifest explaining when and how to load it
• 1,503 stars suggest developers find the modular approach useful
• Targets coding agents, research agents, and workflow agents alike

Caveats

• The README only lists categories; actual depth and quality of individual skills are unclear without inspecting the folders
• It is essentially a curated prompt library, not executable code—value hinges entirely on the quality of the instructions inside

Verdict

Browse it if you’re building agent stacks and want to stop rewriting system prompts from scratch. Skip it if you need frameworks, APIs, or runtime tools.