Best Claude Skills 2026: 15 Skills Worth Installing Today

Claude Skills exploded in 2026. There are now thousands of community-built skills, and 90% of them aren't worth your time. This is a curated list of the 15 that actually punch above their weight — battle-tested, well-described, and useful across real workflows.

All of them are available in our directory and most install in Lovable with one drag-and-drop.

How we picked

Three filters: (1) the skill solves a problem most builders hit, (2) the SKILL.md description is precise enough that Claude actually triggers it, (3) the bundled assets are non-trivial — not just a wrapper around something Claude can already do.

We grouped them by use case.

Foundational (install these first)

1. design-tokens

Defines your color palette, spacing scale, typography, and radii as a single source of truth Claude reads before generating any UI. Eliminates the "why is every button a different shade of blue" problem on day one. View

2. shadcn-blocks

Pre-composed shadcn/ui sections (hero, pricing, FAQ, dashboard shells) Claude pulls in instead of reinventing layout every time. Massive speedup on landing pages and admin UI. View

3. ui-ux-pro-max-v2

Heuristics + checklists for accessible, conversion-friendly interfaces. Catches the small stuff (focus states, error microcopy, empty states) most LLM-generated UI forgets. View

Backend & infrastructure

4. resend-email

Transactional email patterns wired for Resend: templates, retry logic, signed webhooks, bounce handling. Saves a half-day every time you need email in a new project.

5. metrics-dashboard

Drop-in skill for shipping a metrics page: server function patterns, recharts presets, sensible defaults for ranges and aggregations.

6. graphmemory

DuckDB-backed knowledge graph for agents that need persistent, queryable memory across sessions. Ideal for support bots, research agents, and any "remember what we discussed" use case.

AI & agents

7. claude-agent

Templates for spinning up subagents with scoped tools, system prompts, and handoff conventions. Removes the boilerplate when your app needs more than one agent persona. View

8. union-search-skill

Federated search across 20+ engines (web, code, docs, papers, video) with deduped results. The right primitive whenever your agent needs "search the internet" without hardcoding one provider.

9. depth-estimation

Runs depth estimation on images locally in the browser — useful for privacy-respecting AR, scene understanding, and any "I don't want to ship user images to a third party" use case.

DevOps & observability

10. yeet-scripts

Real-time AI + system observability scripts. Tail logs, watch metrics, alert on anomalies — all wired to a small CLI Claude can drive on your behalf.

11. ctct-security-patrol

Web ops hardening skill: dependency audits, header checks, common-vuln scans. Catches the low-hanging security issues before you ship.

12. superpowers-streamer-cli

CLI for orchestrating long-running AI automations with streaming output. Best fit when you have multi-step generation pipelines that take minutes.

Content & marketing

13. brand-voice (pattern, multiple implementations)

Captures your tone of voice, banned words, signature phrases. Once installed, every email, landing page, and tweet Claude writes sounds like you — not like a generic LLM.

14. weekly-report (pattern)

Turns a CSV of metrics into a styled report. The canonical example of "boring task you do every week, now it does itself."

15. immortal-skill

Framework for building digital twins — agents that absorb your writing, decisions, and preferences over time. Great starting point if you want a long-lived, personalized assistant. View

How to pick what to install

Don't install all 15. Start with 2-3 from Foundational so every project benefits, then add domain-specific skills as the work demands.

A reasonable "starter pack" for most builders:

design-tokens (so UI is consistent)
shadcn-blocks (so UI is fast)
One backend skill matching your stack (resend-email, metrics-dashboard, etc.)

Layer in the rest only when you hit the problem they solve. A skill you never trigger is just clutter in Claude's context window.

Installing in Lovable

All skills in our directory install the same way:

Click the skill, hit Download .zip
Drag the zip into a Lovable chat
Lovable detects it, unpacks, and confirms install

For Claude Desktop or Claude Code, unzip into ~/.claude/skills/ and restart.

What's missing from this list

Honest gaps in the current ecosystem (which means opportunity if you build skills):

Payments — no great Stripe-specific skill yet covering subscriptions, webhooks, and customer portal end-to-end
i18n — translation workflows are still mostly DIY
A/B testing — experiment frameworks aren't packaged as skills yet
Compliance — GDPR / SOC2 checklist skills are early

If you build one of these well, it'll likely top this list next year. See our How to Create a Claude Skill guide to get started.

Closing thought

The best skill is the one you'd otherwise re-explain to Claude every week. Look at your last 10 conversations — if you typed the same context paragraph more than twice, bundle it. Then submit it to the directory so the next builder doesn't have to.

Browse the full Claude & Lovable Skills catalog — updated daily.