I tested 200 Claude Code skills so you don't have to — here are the 20 that actually changed how I work

Been deep in the Anthropic Claude skills ecosystem for the past several months. Tested roughly 200 of them. Most are noise. These 20 are worth the time.
Quick context for those unfamiliar: Claude skills are SKILL.md folders that give Claude Code specialized knowledge for specific tasks — essentially SOPs for an AI agent. The right skill loads when relevant, the rest are ignored. The ecosystem just crossed 60,000 published skills. The signal-to-noise ratio is rough. This post is an attempt to fix that.

The filter used
Three criteria only:

• No coding knowledge required to install or use
• Install-and-run (no complex configuration)
• Worth using again after the first session
  If something didn't pass all three, it didn't make the list.

Layer 1: The document skills (start here)
Anthropic's official four are the most universally useful claude code skills available. The skepticism is understandable — "AI does Word docs" sounds like a press release. But the PDF skill alone changed contract review workflows meaningfully.
A 48-page agreement that used to take 90 minutes of careful reading now takes under a minute. Risk points flagged, clauses extracted, ambiguous language highlighted. One caveat: scanned PDFs (image-based) need OCR processing first before the skill can work effectively.

• PDF → github.com/anthropics/skills/tree/main/skills/pdf
• DOCX → github.com/anthropics/skills/tree/main/skills/docx
• PPTX → github.com/anthropics/skills/tree/main/skills/pptx
• XLSX → github.com/anthropics/skills/tree/main/skills/xlsx
  If you take nothing else from this post, install these four. They cover the majority of what most knowledge workers actually spend time on.

Layer 2: Where it gets more interesting
Frontend Design (github.com/anthropics/skills/tree/main/skills/frontend-design)
AI-generated UIs have a well-earned reputation for looking generic — cramped layouts, poor color choices, no visual hierarchy. The industry calls it "AI slop." This official Anthropic claude skill embeds a design system that systematically raises the floor on generated interfaces. Same prompt, noticeably better output. Still not replacing a designer for anything client-facing, but internal tools and prototypes become shareable without embarrassment.
Marketing Skills (github.com/coreyhaines31/marketingskills)
Built by Corey Haines specifically for technical founders. Twenty-plus claude code skills covering CRO, copywriting, SEO, email sequences, A/B testing, and growth strategy — all in one install. For solo founders wearing too many hats, this is the one collection worth having rather than hunting individual skills across repositories.
Claude SEO (github.com/AgriciDaniel/claude-seo)
Drop a URL, receive a structured audit: technical SEO issues, page-level analysis, schema markup validation, sitemap generation, and AI search optimization (GEO/AEO). Structured as 12 sub-skills running parallel subagents for thoroughness. For anyone running a content site or SaaS with organic traffic ambitions, this makes regular audits practical rather than aspirational.
NotebookLM Skill (github.com/PleasePrompto/notebooklm-skill)
The most underrated entry on the list. This claude agent skill bridges Claude Code and Google NotebookLM — automatically querying uploaded documents and returning source-grounded summaries, flashcards, and research notes. A 30-page technical paper that used to require 3-4 hours of careful processing now takes about 15 minutes. The flashcards it produces are genuinely evaluative rather than simple recall prompts.
Two requirements worth noting: a Google account with NotebookLM access, and local Claude Code (the web UI won't work here due to sandbox restrictions on browser automation).
Obsidian Skills (github.com/kepano/obsidian-skills)
Built by Obsidian's own CEO. If Obsidian is part of the knowledge management workflow, this is the skills collection to add. Claude starts creating notes that conform to proper vault structure — auto-tagging, auto-linking, Markdown formatted correctly. The second brain starts maintaining itself.
Amazon Competitor Analyzer (github.com/browser-act/skills/blob/main/amazon-competitor-analyzer/SKILL.md)
Comprehensive competitive intelligence for Amazon sellers and e-commerce teams. Provide a product ASIN or search query, and this claude skill extracts pricing, ratings, review counts, and market positioning from Amazon — without getting blocked by CAPTCHA or anti-bot systems, thanks to BrowserAct's browser automation layer. For anyone doing e-commerce market research manually today, this turns a multi-hour process into a single prompt.
Skill Prompt Generator (github.com/huangserva/skill-prompt-generator)
Text-to-image prompt engineering is its own discipline. This claude skill takes a described visual concept and produces optimized, professional prompt language for image generation tools. Eliminates the trial-and-error loop that normally eats up an hour before arriving at a prompt that actually works.

Layer 3: The one that changes everything
Skill Creator (github.com/anthropics/skills/tree/main/skills/skill-creator)
This is Anthropic's meta-skill — a claude skill for building other claude skills. And it's the most important entry on this list by a significant margin.
Here's the problem it solves: professional expertise accumulates over years. Workflows, formatting preferences, editorial standards, judgment calls about edge cases. Every new Claude conversation starts completely cold. All of that context gets re-explained from scratch, every single time.
Skill-creator externalizes that institutional knowledge into a reusable SKILL.md file that persists permanently.
The workflow is straightforward: describe the process in plain language, answer five or six clarifying questions about audience, trigger conditions, and output format, and receive a complete claude skill specification in roughly five minutes. No code required at any stage.
The result: a Claude agent that already knows how to work in a specific style — without re-briefing. Built a content workflow skill from scratch using it recently. Eight minutes total. Now Claude knows the formatting preferences, section structure, and editorial standards without restating them every session.
This is where the real divergence happens. The majority of claude code skill users consume what others built. The smaller group that builds their own — encoding actual domain expertise into persistent, reusable skills — is building something that compounds over time. Skill-creator is what makes that accessible to non-developers.
Superpowers (github.com/obra/superpowers)
Standard Claude plan mode moves from request to execution immediately. Superpowers inserts a structured phase first: systematic requirement interrogation, use case clarification, edge case identification, and explicit implementation planning before any code or content is produced.
The analogy is a senior consultant running a requirements workshop versus a junior developer who starts building immediately upon hearing the request. For complex projects, the difference in final output quality is substantial. Superpowers packages 20+ battle-tested claude code skills covering planning methodology, test-driven development, systematic debugging, and collaboration patterns.

Full writeup with context on each skill: https://www.browseract.com/blog/best-claude-skills