Teaching Claude to stop winging it

What it does This repository is an awesome-list that catalogs Claude Skills—specialized folders of instructions, scripts, and resources that Claude Code (and other Claude interfaces) dynamically loads when relevant. It maps both Anthropic’s official offerings and community-built skills, from Excel manipulation to iOS automation, plus tools for authoring your own.

The interesting bit The list itself is glue, but the architecture it describes is clever: skills use progressive disclosure, exposing only ~100 tokens of metadata for discovery until Claude decides the full instruction set is actually needed. That keeps dozens of skills available without clogging the context window.

Key highlights

• Official skills cover document heavy lifting (docx, pdf, pptx, xlsx), frontend design, Playwright testing, and MCP server creation.
• Community collections like obra/superpowers bundle 20+ battle-tested skills with built-in commands such as /brainstorm and /execute-plan.
• One entry, loki-mode, ambitiously orchestrates 37 AI agents across six swarms to run a startup from spec to deployment.
• The repo points to skill-creator, an official interactive tool that generates new skills through a guided Q&A.
• A warning is plainly posted: skills can execute arbitrary code in Claude’s environment, so vet before you install.

Caveats

• The community skills section is currently a flat, uncategorized table; the maintainers note it will be organized once the volume justifies it.
• Skills execute arbitrary code, making security auditing your own responsibility.

Verdict Worth bookmarking if you use Claude Code daily and want repeatable, specialized behavior without hand-crafting prompts. If you don’t use Claude, it’s just a very well-organized list of things you can’t use.