Why this page
Claude is only as good as what you hand it. On its own it works from memory, and memory goes stale. A skill gives it a checklist. An MCP server gives it the real documentation and the real system.
The difference in output is large. Do not skip this and then complain the answers are wrong.
Install everything
Run these in a terminal. You only need the marketplace lines once.
# Marketplaces
claude plugin marketplace add cloudflare/skills
claude plugin marketplace add sveltejs/ai-tools
# Everyone
claude plugin install code-review@claude-plugins-official
claude plugin install context7@claude-plugins-official
claude plugin install feature-dev@claude-plugins-official
claude plugin install code-simplifier@claude-plugins-official
claude plugin install planning-with-files@claude-plugins-official
# Web and frontend
claude plugin install playwright@claude-plugins-official
claude plugin install frontend-design@claude-plugins-official
claude plugin install typescript-lsp@claude-plugins-official
claude plugin install svelte@svelte
claude plugin install cloudflare@cloudflare
# .NET
claude plugin install csharp-lsp@claude-plugins-official
# Shopify
claude plugin install shopify-ai-toolkit@claude-plugins-officialInside a session use /plugin install <name>@<marketplace> instead. /plugin on
its own opens the browser.
claude plugin list shows what you already have.
Required for everyone
Three things. Not optional.
| Tool | What it does | When |
|---|---|---|
/code-review |
Reviews your changes for bugs, security holes, and code that does not match the project | After every bug fix and every feature, before the pull request |
find-docs |
Pulls the current documentation for any library or framework | Before you use any API you have not used this week |
git-guardrails-claude-code |
Blocks force push, hard reset, branch delete, and clean before they run | Always on. It costs nothing until the day it saves you |
/code-review ultra is the bigger version. It runs many reviewers in parallel
over the whole branch. Use it before a release, or on anything touching money,
login, or customer data.
Most days
| Skill | What it does |
|---|---|
simplify |
Cleans up what you just wrote. Cuts repetition and reuses what already exists |
diagnosing-bugs |
A real diagnosis loop for bugs with no obvious cause |
implement |
Works through a feature end to end |
planning-with-files |
Plans the work in files first, so the plan survives a long session |
find-skills |
Finds a skill for whatever you are stuck on |
triage |
Sorts a pile of issues into what matters |
When they fit
| Skill | Use it when |
|---|---|
tdd |
You want the test written before the code |
expect |
The change is visible in a browser and needs testing against a real page |
qa |
You are reporting bugs and want them filed as you talk |
prototype |
You need to answer a design question before committing to it |
setup-pre-commit |
A repo has no formatting or type checks before commit |
resolving-merge-conflicts |
You are stuck in a merge or rebase |
request-refactor-plan |
A refactor needs breaking into safe steps |
improve-codebase-architecture |
The structure itself is the problem |
codebase-design |
You are designing a module interface |
design-an-interface |
You want several API designs to compare |
to-spec, to-tickets |
Turning a conversation into a spec or into issues |
By stack
Web, Astro, Svelte
| Tool | For |
|---|---|
svelte plugin |
Svelte 5 docs, code validation, and a Svelte file editor agent |
typescript-lsp |
Real TypeScript errors instead of guesses |
modern-web-guidance |
Current HTML and CSS practice. Run it first on any UI task |
web-design-guidelines |
Checks a UI for accessibility and layout problems |
frontend-design plugin |
Visual direction for new UI |
design-taste-frontend |
Landing pages and redesigns that do not look templated |
taste-skill |
Extracts a design system from any site by URL |
shadcn |
shadcn/ui components and registries |
vercel-react-best-practices |
React and Next.js performance |
next-partial-prefetching-adoption |
Turning on Next.js partial prefetching |
Cloudflare
The cloudflare plugin brings five MCP servers and the Workers skills. Use it
for Workers, Pages, D1, R2, KV, Agents SDK, and Wrangler. It reads real build and
runtime logs, which is the part that matters when a deploy fails.
.NET, WPF, MAUI
| Tool | For |
|---|---|
csharp-lsp plugin |
Real C# errors. Turn this on before touching DawloomAttendance |
wpf-expert |
WPF work |
tw-maui |
Styling .NET MAUI with Tailwind classes through Tw.Maui |
POS
pos-feature-dev scaffolds a feature module for the POS SaaS in the right order:
Prisma, schemas, tRPC, table and modal, page, sidebar.
Mobile and Expo
| Skill | For |
|---|---|
building-native-ui |
Expo Router screens, navigation, styling |
expo-deployment |
Shipping to the App Store, Play Store, and web |
upgrading-expo |
SDK upgrades and the dependency mess that follows |
Shopify
shopify-ai-toolkit covers Admin GraphQL, Liquid themes, Hydrogen, Functions,
Polaris, checkout and admin extensions, and the Shopify CLI.
Video and media
remotion-best-practices for video built in React. video-use for editing an
existing video by conversation.
Writing
For anything a client or the team reads.
| Skill | For |
|---|---|
human-writing |
Docs, READMEs, PRDs, emails. Removes the AI tone |
humanizer |
Strips AI writing patterns out of existing text |
research |
Investigates a question against real sources and writes it up |
MCP servers
An MCP server lets Claude read the real system instead of guessing.
| Server | Comes from | Gives Claude |
|---|---|---|
| Context7 | context7 plugin |
Current docs for almost any library. This is what find-docs uses |
| Cloudflare | cloudflare plugin |
Workers, bindings, builds, docs, and observability. Five servers |
| Playwright | playwright plugin |
A real browser it can drive and inspect |
| Svelte | svelte plugin |
Svelte 5 docs and code validation |
| Supabase | claude.ai account | Your actual tables, migrations, and advisors |
| Vercel | claude.ai account | Deployments, build logs, runtime errors |
| Claude in Chrome | built in | Your own browser, for testing a real logged-in session |
Account connectors are added at claude.ai under connectors, not by a terminal command.
Adding skills
Skills are plain folders in ~/.claude/skills/. Drop one in and it loads next
session.
To find one, use the find-skills skill and ask in plain words, for example
“is there a skill for writing Expo screens”.
To write one for a pattern your team repeats, run claude plugin init <name>.
Before you integrate anything
Check the official documentation of any library or service for an MCP server or an agent skill before you write code against it. Most serious projects ship one now.
If there is none, run find-docs first. See the
using AI to write code page.
Examples
You just fixed a bug and the tests pass. Are you done?
No. Run /code-review first.
Passing tests prove the case you thought of works. The review is what catches the case you did not think of.
You are adding Stripe to a project for the first time.
Two steps before any code.
Check Stripe’s official docs for an MCP server or an agent skill. Then run find-docs for the API you are about to call.
Both take a minute. Guessing at a payments API costs far more.
Claude wrote code using a method that does not exist.
It worked from memory. That is what memory does.
Run find-docs for that library and ask again. If it happens twice with the same library, turn on its MCP server.
You are about to change the C# attendance app.
Install csharp-lsp first.
Without it Claude cannot see real compiler errors and will confidently write code that does not build.
A Cloudflare deploy failed and the error is not obvious.
The cloudflare plugin can read the actual build and runtime logs.
Pasting a screenshot of the dashboard gives it far less to work with.
If something here is wrong
If this does not match how we really work, or you think it should change, email support@dawloom.com.