Teaching Claude specialized trades without the token tax

What it does

This repository catalogs over 50 “Agent Skills” for Claude—modular instruction packs that nudge the model toward specific workflows like test-driven development, PDF extraction, or systematic debugging. Each skill lives in its own folder and loads only when relevant, so Claude isn’t permanently saddled with knowledge it isn’t using. It functions as an opinionated awesome-list bridging official Anthropic skills and community contributions.

The interesting bit

The README makes a pragmatic case for skills over system prompts or MCP servers: they cost 30–50 tokens until loaded, activate automatically based on context, and work across Claude Code, the web app, and the API. That’s a rare focus on token economics and cross-platform portability in a space that usually just throws more context at the problem.

Key highlights

• 50+ skills across a dozen categories, from document processing to debugging and git workflows
• Mix of verified community skills (notably obra/superpowers) and official Anthropic offerings
• Automatic, context-aware activation rather than always-on prompt bloat
• Explicit comparison table weighing Skills against MCP servers and system prompts
• Some entries marked “community-needed” for gaps like end-to-end testing and performance profiling

Caveats

• Several listed skills are marked “community-needed” but not yet filled, so the catalog has aspirational gaps
• “Verified” status appears to be a community judgment call rather than an Anthropic seal of approval
• Requires Claude Pro, Max, Team, or Enterprise plans with code execution enabled

Verdict

Worth bookmarking if you use Claude Code daily and keep repeating the same prompts; skip it if you expect fully original research—this is aggregation and curation, not invention.