In 2026, Claude Code has evolved from a chatty terminal into an orchestrable Agent platform. What often separates teams is not the model version—it is whether you have installed the right Skills.
Think of a Skill as an operations manual for your AI: when you type a slash command like /grill-me, Claude follows the manual—clarify requirements before coding, write tests before implementation, break large requests into small tasks—instead of changing code wherever inspiration strikes.
The problem: skills.sh already lists thousands of skills, many with English names. This article explains in plain language what each Skill does and whether you should install it, then provides one-click install commands. For a deeper dive into the mattpocock single-pack design philosophy, see our mattpocock/skills deep dive.
Every Skill below includes a copy-paste npx skills add ... command matching the install command on each skills.sh skill card—paste directly into your terminal.
1. What Skills are, and how they differ from Rules
In one line: Skills are on-demand workflows; Rules are always-on project conventions.
An analogy: Rules are like company attendance policy—everyone follows them every day. Skills are SOPs for writing requirements, running tests, or doing code review—you open the right one only when you need that task. Technically, each Skill is a folder with a SKILL.md that tells Claude the steps; you trigger it by typing /skill-name in the conversation.
| Mechanism | Plain-language meaning | When it applies |
|---|---|---|
| Rules / CLAUDE.md | Project "house rules": code style, directory layout, forbidden actions | Included in almost every conversation |
| Skills | Reusable "how-to" workflows: align requirements, write tests, break down tasks | When you type /xxx or Claude decides one is needed |
| Claude Code built-in skills | Official defaults: code review, debug loops | Type /code-review, /debug, and similar |
How to combine them: Rules set the floor; Skills run the process. For example, a Rule might say "no force push"; Skill /git-guardrails-claude-code blocks dangerous Git commands at execution time.
2. One-click install: copy commands to your terminal
You do not need to download files manually. The community standard is the skills CLI (npx skills): it syncs GitHub skill packs to your Claude Code directory (Cursor and Codex are supported too).
On any skill card at skills.sh, the page shows a copyable npx skills add ... command—commands in this article match the official site; copy and paste into your terminal to install.
2.1 General command format
# Install a single Skill (global + skip confirmation + target Claude Code)
npx skills add owner/repo@skill-name -g -y -a claude-code
# Install an entire skill pack (interactive selection, or add --all)
npx skills add owner/repo -g -y -a claude-code
# Install to current project only (team can commit .claude/skills/)
npx skills add owner/repo@skill-name -y -a claude-code2.2 Key flags at a glance
| Flag | Meaning |
|---|---|
-g, --global | Install to ~/.claude/skills/, available in all projects |
-y, --yes | Skip confirmation, good for scripted one-click installs |
-a claude-code | Write only to Claude Code directory (can add cursor, codex) |
-s skill-name | Install only the named Skill from a large pack |
--copy | Copy files instead of symlink (needed in some environments) |
3. Tier 0: Meta-skills (install these two first)
This tier does not write code for you—it helps you find and create skills. Install these first so you do not have to keep returning to this article.
find-skills — the Skills app store
What it does: Search and install skills from skills.sh directly inside Claude Code. Like browsing an app store and installing without switching to a browser to copy commands.
When to use it: You want skills for security audits, SEO, databases, etc., but do not know the repo names; or you want to refresh your stack every few months.
One-click install: Copy the command below to your terminal (matches skills.sh).
npx skills add vercel-labs/skills@find-skills -g -y -a claude-codeskill-creator — scaffold for writing your own Skill
What it does: Guides you through writing a team-specific Skill in Anthropic's official format (release checklist, code review checklist, etc.). Ship it internally or open-source on GitHub.
When to use it: You find yourself repeating the same instructions to Claude every week and want to codify them as a one-shot command like /our-deploy.
One-click install: Copy the command below to your terminal.
npx skills add anthropics/skills@skill-creator -g -y -a claude-code4. Tier 1: Engineering discipline pack (mattpocock/skills)
mattpocock/skills is a six-figure-star skill pack on GitHub. Matt Pocock (creator of Total TypeScript) open-sourced the workflows he uses daily. Core value: stop AI from coding before it understands, shipping without tests, or touching fifty files at once.
If you install only one third-party pack, start here. Below, each skill is listed in "requirements to delivery" order—understand the purpose before deciding whether to install individually.
Install the full pack
One-click full pack: Copy the command below; after install, run /setup-matt-pocock-skills to configure the repo.
npx skills add mattpocock/skills -g -y -a claude-codeAfter install, you must run /setup-matt-pocock-skills at the repo root: tell Claude whether you use GitHub or Linear for tasks, where team terminology lives, and which directory holds docs. Skip this and issue breakdown may not match how your team works.
4.1 Requirements alignment — clarify before coding
| Command | What it does | Typical scenario |
|---|---|---|
/grill-me |
Like a senior colleague interrogating requirements: edge cases, failure modes, acceptance criteria. Claude will not touch code until both sides align. | New feature or core logic change when the requirement is still one sentence |
/grill-with-docs |
While questioning, write consensus into CONTEXT.md and ADRs so future conversations can reference team shorthand. |
When you want AI to remember "we call this XXX" terminology |
/to-prd |
Turn clarified discussion into a PRD and open a GitHub Issue automatically. | After aligning with product or leadership, need written output for the team |
/to-issues |
Split a PRD into multiple independently shippable vertical Issues, avoiding "frontend week, backend week" horizontal slices. | Planning sprints, assigning work, or having AI implement issue by issue |
4.2 Coding and quality — verifiable, reversible changes
| Command | What it does | Typical scenario |
|---|---|---|
/tdd |
Write failing test → minimal code to pass → refactor. Stops AI from dumping hundreds of lines that do not run. | Business logic, bug fixes, modules where unit tests are feasible |
/diagnose |
Structured debugging: reproduce → narrow scope → hypothesize → verify → fix → regression. Less "try a change and see." | Intermittent bugs, performance, concurrency that are not obvious at a glance |
/improve-codebase-architecture |
Periodically scan the codebase for blurry module boundaries, inconsistent naming, things that should be split, and suggest refactors. | Before a release, or every week or two for an "architecture health check" |
/git-guardrails-claude-code |
Gate Git commands: block force push, hard reset, and other dangerous operations before AI slips and deletes history. | Repos with production branches; teams with "AI blew up main" scars |
4.3 Efficiency and collaboration
| Command | What it does | Typical scenario |
|---|---|---|
/caveman |
Makes Claude reply in ultra-short sentences, less fluff and repetition—roughly 75% token savings. | Long sessions, cost-sensitive work, or when you want results not essays |
/handoff |
Compress conversation context, progress, and open items into a handoff document for another person or Agent to continue. | Stop halfway today, resume tomorrow, or switching from Claude Code to Cursor |
/setup-matt-pocock-skills |
First-run setup wizard: pick issue platform, label rules, doc paths. Required after full pack install. | Right after installing mattpocock/skills |
Recommended daily chain: new feature → /grill-with-docs to align → /to-issues to split tasks → /tdd per task to implement. For a deeper walkthrough, see our mattpocock deep dive.
Single-skill install: Need only one? Copy the matching line below (format matches skills.sh skill cards).
npx skills add mattpocock/skills@grill-me -g -y -a claude-code
npx skills add mattpocock/skills@tdd -g -y -a claude-code
npx skills add mattpocock/skills@to-issues -g -y -a claude-code5. Tier 2: Sub-agent orchestration (obra/superpowers)
obra/superpowers solves a different problem: the task is too big for one Agent chatting start to finish. It teaches plan first, delegate to sub-agents in parallel, then integrate—like a project manager leading a team, not one person doing everything.
Division of labor with mattpocock: mattpocock handles daily coding discipline; superpowers handles how to split and parallelize large requests. You can install both.
Install the full pack
One-click install: Copy the command below to your terminal.
npx skills add obra/superpowers -g -y -a claude-code| Command | What it does | Typical scenario |
|---|---|---|
/brainstorming |
Brainstorm large requests first: list options, weigh pros and cons, assess risks before coding. Stops "build an admin panel" from meaning code immediately. | Vague requirements, multiple tech choices, or multi-module greenfield projects |
/writing-plans |
Turn the chosen approach into a step-by-step execution plan: what each step does, dependencies, acceptance criteria. | Approach is set; you need a trackable task list |
/subagent-driven-development |
Assign sub-tasks to multiple sub-agents in parallel (frontend, API, tests) while the main Agent coordinates merges. | Large surface area, independent modules, want shorter total time |
/test-driven-development |
Another TDD workflow, similar in spirit to mattpocock's /tdd—pick one or mix as you prefer. |
When you want tests to drive implementation from the start |
/systematic-debugging |
Collect evidence → rule out hypotheses → find root cause, instead of "change a line, try again." | Complex bugs, production incidents, cross-service failures |
/verification-before-completion |
Before Claude says "done," it must run tests and pass a checklist. Cures premature victory declarations. | Any task where you do not trust AI to close out on its own |
/requesting-code-review |
Generate a review request for teammates: what changed, why, what to focus on. | Before opening a PR, or when you want AI self-check before human review |
If your team already has mature Spec documentation workflows, you may not need the full pack—installing only /verification-before-completion is often enough.
6. Tier 3: Frontend and design (Anthropic + Vercel)
Many complain that AI-built pages "run but look rough, not like a product." This tier constrains visuals, interaction, and framework usage so output is closer to shippable.
6.1 Anthropic official skills
| Skill | What it does | When to install |
|---|---|---|
frontend-design |
Generate pages to professional UI standards: color, spacing, type hierarchy, component states—less "programmer aesthetic." | Web frontend, landing pages, dashboards when visual quality matters |
webapp-testing |
Guides Claude to actually open and test the page: click buttons, fill forms, read errors—not "should be fine." | After UI changes, want an automated smoke pass |
pdf / docx / pptx / xlsx |
Read and write Office and PDF: reports, contracts, spreadsheets per format rules. | Projects that often handle documents, reports, decks |
One-click install: Copy the commands below (Anthropic official skill pack).
npx skills add anthropics/skills@frontend-design -g -y -a claude-code
npx skills add anthropics/skills@webapp-testing -g -y -a claude-code
npx skills add anthropics/skills@pdf -g -y -a claude-code6.2 Vercel stack: how to write React / Next.js correctly
| Skill | What it does | When to install |
|---|---|---|
vercel-react-best-practices |
Code to Vercel team's React best practices: component split, state, performance, App Router, etc. | React / Next.js projects when you want AI to avoid outdated patterns |
web-design-guidelines |
Check against web design standards: accessibility, responsive layout, contrast, interaction feedback. | Pre-launch UI self-check or design-to-code acceptance |
vercel-react-native-skills |
React Native / Expo writing constraints and common pitfalls. | Cross-platform apps, Expo stack |
deploy-to-vercel |
Step-by-step Vercel deploy: env vars, build command, domain setup. | Project on Vercel when you want AI to walk through deployment |
One-click install: Copy the commands below (Vercel engineering best-practices pack).
npx skills add vercel-labs/agent-skills@vercel-react-best-practices -g -y -a claude-code
npx skills add vercel-labs/agent-skills@web-design-guidelines -g -y -a claude-code6.3 Browser automation — make AI actually "open the page"
agent-browser and playwright-cli let Claude control a real browser: open pages, click, screenshot, capture console errors. Good for E2E tests and "I changed the UI—click through it for me."
One-click install: Copy the commands below (browser automation / E2E testing).
npx skills add vercel-labs/agent-browser -g -y -a claude-code
npx skills add microsoft/playwright-cli -g -y -a claude-codeUsing shadcn/ui? Optional add-on: npx skills add shadcn/ui@shadcn -g -y -a claude-code (copy to terminal and run).
7. Tier 4: Stack-specific picks
Not everyone needs this tier—install what matches your stack so your slash list is not full of unused skills.
| Your scenario | Skill | What it does |
|---|---|---|
| Supabase / Postgres backend | supabase-postgres-best-practices |
Write SQL, RLS policies, indexes, migrations per Supabase recommendations—fewer security and performance traps |
| Firebase full stack | firebase-basics etc. |
Standard patterns for Auth, Hosting, Firestore and other Firebase services |
| Convex realtime backend | convex-quickstart |
Convex schema, mutations, queries, deployment patterns |
| Roll-your-own auth | better-auth-best-practices |
better-auth sessions, OAuth integration best practices |
| Scrape pages / collect data | firecrawl |
Use Firecrawl to crawl and parse pages—data collection and RAG knowledge bases |
| Production error tracking | sentry-cli |
Configure Sentry, inspect errors, tie to releases—helps AI understand your monitoring setup |
| Marketing pages / SEO | seo-audit |
Check titles, meta, page structure against basic SEO requirements |
Example install: Copy the commands below (pick what matches your stack).
npx skills add supabase/agent-skills@supabase-postgres-best-practices -g -y -a claude-code
npx skills add firecrawl/cli@firecrawl -g -y -a claude-codeFor native iOS development, Claude Code does not replace Xcode; for system-side Agent changes see WWDC26: Xcode 27 Agent and the iOS Development Shift—builds should still run xcodebuild on macOS.
8. Three recommended bundles (copy and run)
If you do not want to pick one by one, choose a bundle by role. Each includes a short why this combination.
8.1 Solo developer · minimum must-haves
Who it is for: Indie devs, small projects, new to Claude Code.
Logic: Discover more skills + align before coding + test before implementation + pages that are not ugly.
Copy the whole block to your terminal: You can paste all commands at once (same format as single installs on skills.sh).
npx skills add vercel-labs/skills@find-skills -g -y -a claude-code
npx skills add mattpocock/skills@grill-me -g -y -a claude-code
npx skills add mattpocock/skills@tdd -g -y -a claude-code
npx skills add mattpocock/skills@setup-matt-pocock-skills -g -y -a claude-code
npx skills add anthropics/skills@frontend-design -g -y -a claude-code8.2 Full-stack engineer · standard kit
Who it is for: Daily React/Next.js work, medium-complexity tasks.
Logic: Full mattpocock daily discipline + superpowers against premature "done" + Vercel frontend standards + browser testing.
Copy the whole block to your terminal: You can paste all commands at once.
npx skills add vercel-labs/skills@find-skills -g -y -a claude-code
npx skills add mattpocock/skills -g -y -a claude-code
npx skills add obra/superpowers@brainstorming -g -y -a claude-code
npx skills add obra/superpowers@verification-before-completion -g -y -a claude-code
npx skills add vercel-labs/agent-skills@vercel-react-best-practices -g -y -a claude-code
npx skills add vercel-labs/agent-skills@web-design-guidelines -g -y -a claude-code
npx skills add anthropics/skills@webapp-testing -g -y -a claude-code8.3 Tech lead · discipline + orchestration + safety
Who it is for: Leading a team, standardizing AI collaboration, worried about Git accidents.
Logic: Full engineering discipline pack + full orchestration pack + Git guardrails + ability to author team Skills.
Copy the whole block to your terminal: You can paste all commands at once.
npx skills add mattpocock/skills -g -y -a claude-code
npx skills add obra/superpowers -g -y -a claude-code
npx skills add mattpocock/skills@git-guardrails-claude-code -g -y -a claude-code
npx skills add anthropics/skills@skill-creator -g -y -a claude-code
npx skills add mattpocock/skills@write-a-skill -g -y -a claude-codeIf your team also uses Cursor: change -a claude-code to -a claude-code,cursor in the commands—one paste writes to both tools.
9. How to verify after installing
- Check the directory: Global Skills should appear under
~/.claude/skills/with a folder containingSKILL.md. - Slash completion: In a new Claude Code session, type
/—installed skill names (e.g.grill-me,tdd) should appear. - Run one real chain: On a small feature branch, run
/grill-meand confirm it asks questions instead of editing code immediately. - mattpocock pack: Run
/setup-matt-pocock-skillsand confirm team CONTEXT paths are created as expected. - Supply chain: Install only Skills from traceable GitHub sources; for unfamiliar repos, check stars, commit history, and whether
SKILL.mdasks you to run unknown scripts.
If you connect Claude Code to local Ollama to save API costs, Skills work the same—process constraints are model-agnostic. Setup steps are in Claude Code + Ollama: Local Agent Workflow on M4 Mac Mini.
10. Anti-patterns: when not to install everything
- Hotfixes under ~50 lines: State acceptance criteria clearly; no need for the full
/grill-meflow. - Spec-Kit or internal process already works: Cherry-pick
/tdd,/diagnose, etc.—avoid stacking redundant process layers. - No one maintains CONTEXT.md: Assign a doc owner before enabling
/grill-with-docs. - Installing 50+ Skills you never use: Clutters the slash list and raises the chance Claude picks the wrong workflow.
- Core need is IM bots / Cron: Prioritize OpenClaw-style gateway Skills over the mattpocock engineering pack.
Leaderboard install counts mean "many people tried it," not "your repo needs it today." Use find-skills or skills.sh quarterly to scan for new skills—no need to chase the entire leaderboard.
FAQ
What is the difference between Claude Code Skills and Cursor Rules?
Rules are "house rules" included in almost every conversation; Skills are "how-to workflows" you invoke with /commands when needed. Analogy: a Rule says "code must have tests"; Skill /tdd walks you through failing test first, then implementation.
Do I need to restart Claude Code after installing a Skill?
After npx skills add, changes usually take effect immediately; if you manually copy folders to ~/.claude/skills/, exit and reopen your Claude Code session.
What is the difference between -g and not using -g?
-g installs globally to ~/.claude/skills/ (all projects); without -g, install goes to the current directory's .claude/skills/, suitable for team sharing via Git.
What are the three Skills worth installing first?
Most developers can start with find-skills, grill-me (or grill-with-docs), and tdd; for large requests add superpowers' brainstorming and verification-before-completion.
Can Skills be used in Cursor?
Yes. Add -a cursor or -a claude-code,cursor when installing—one npx command writes to both Cursor and Claude Code.
Will installing too many Skills slow down Claude Code?
Skill metadata enters the discovery list, but full instructions load only when triggered. Install in tiers, validate each batch with a real task, and avoid installing dozens of skills you never use.
Conclusion
In the 2026 Claude Code ecosystem, Skills are the lowest-cost lever to turn "can write code" into "writes code with engineering discipline." skills.sh makes installation a single copied npx skills add line; this article organizes five tiers—meta-skills → engineering discipline → sub-agent orchestration → frontend quality → stack-specific—and provides three paste-ready bundle scripts.
- Everyone should install:
find-skills+ mattpocock'sgrill-me/tdd+setup-matt-pocock-skills. - Add for large requests: obra/superpowers'
brainstorming,subagent-driven-development,verification-before-completion. - Add for UI work:
anthropics/skills@frontend-design+ Vercel React best-practices pack. - Maintenance: Refresh with
find-skillsor the skills.sh leaderboard—do not hoard unused skills in one shot.
Skills standardize how you collaborate; if you also need a stable macOS environment for iOS builds or 24×7 Agents, visit the Macstripe home page for dedicated M4 Mac Mini nodes. More AI coding practice is in the Developer Blog index.
Related reading
- GitHub Hit: mattpocock/skills Redefines How We Collaborate with AI Coding Agents
- How to Set Up a Local AI Agent on an M4 Mac Mini? One-Month ~80% API Cost Savings (Real Test)
- WWDC26 Drops Xcode 27 Agent: The IDE Just Changed Forever
- Cursor Misses the Frontend? CodeGraph Fix for Vue + Spring Boot
- 2026 AI Coding Cost Rankings: Claude, Codex, Cursor, Gemini — Which Burns the Most?
