The 7 Claude Code plugins we keep installed

Claude Code’s plugin ecosystem grew fast. Too fast. There are now over 140+ official plugins in the marketplace, ranging from deeply useful to “nice demo, never again.” After testing 80+ of them, seven earned a permanent place in every project: Skill Creator, Frontend Design, Playwright, CLAUDE.md Management, Code Review, Context7, and GitHub or GitLab. This post covers what each one does, the install count behind it, and why it survived when most of the marketplace did not.

We installed, tested, and uninstalled a lot of plugins over the past months. Most are situational, and great if you happen to be on Supabase, or deploying to Railway, or working in a specific language. But seven of them we kept using across every project. These are the ones that changed how we actually work, not just what we can demo.

Here’s the list, ordered by how they fit into a typical development workflow.

1. Skill Creator: teach Claude your team’s workflows

176k installs · claude.com/plugins/skill-creator

Claude Code ships with powerful defaults, but every team has workflows that don’t map neatly onto generic commands. Skill Creator lets you encode those workflows as reusable slash commands. Your deployment checklist becomes /deploy. Your PR description template becomes /pr. Your security review process becomes /sec-review.

This is the meta plugin, the one that makes Claude Code yours. You define what a skill does, Skill Creator scaffolds the implementation, and from that point on the workflow is repeatable by anyone on the team. It also handles evaluating and benchmarking skills, so you can measure whether a skill is actually doing what you intended before rolling it out. The same progressive disclosure idea shows up in agent skills elsewhere.

If you’ve ever found yourself repeating the same multi-step instruction to Claude across different sessions, that instruction belongs in a skill.

2. Frontend Design: UI that doesn’t look AI-generated

564k installs · claude.com/plugins/frontend-design

The most-downloaded plugin in the entire marketplace, and it earns it. The generic problem with AI-generated UI is that it looks like AI-generated UI: overly symmetric, generic color palettes, no visual hierarchy. Frontend Design specifically addresses this.

It gives Claude specialized knowledge of production design patterns, component library idioms, visual rhythm, and spacing systems. The generated code tends toward components that look like they came from a design system, not a template. Particularly useful when you’re building something client-facing and don’t want to spend cycles redesigning every component after generation.

Reality check: this plugin raises the floor of AI-generated UI significantly. It doesn’t replace a designer. It makes the starting point defensible enough to iterate from rather than rebuild from scratch.

3. Playwright: browser access for Claude

180k installs · claude.com/plugins/playwright

This one changes what Claude Code can fundamentally do. Neither the terminal nor the desktop app gives Claude access to a browser (yet). Without Playwright, frontend work is blind: Claude writes the code, you manually verify it, you describe what’s wrong, Claude guesses at fixes.

With Playwright, Claude can navigate to a URL, take screenshots, interact with forms, assert on DOM content, and catch visual states, all within the same conversation. The feedback loop for frontend iteration collapses. You describe what should happen, Claude implements it, tests it in the browser, and verifies the result without you switching windows.

For any project with a frontend component, this is a force multiplier. It pairs especially well with Frontend Design.

4. CLAUDE.md Management: keep your project memory accurate

155k installs · claude.com/plugins/claude-md-management

Your CLAUDE.md file is what gives Claude persistent context about your project: architecture, conventions, key paths, preferred patterns. It starts accurate and gradually drifts as projects evolve, through new dependencies, renamed files, and refactored patterns. It is the closest thing a coding agent has to durable project memory.

CLAUDE.md Management audits your file against the current state of the codebase and surfaces what’s stale or missing. It doesn’t rewrite it blindly; it captures learnings from recent sessions and proposes specific updates. The result is a CLAUDE.md that stays synchronized with your actual project rather than the project you had six months ago.

At scale, this has a compounding effect: the more accurate your project context, the more accurate every Claude response that relies on it.

5. Code Review: specialized agents with confidence filtering

255k installs · claude.com/plugins/code-review

Generic code review from a single model prompt tends to be either too cautious (“looks okay to me”) or too noisy, flagging stylistic preferences as bugs. Code Review runs specialized agents for security, logic, types, and quality, each focused on a specific domain, with confidence-based filtering that suppresses low-certainty flags.

The result is pull request feedback that reads like it came from someone who actually knows what they’re looking for, not a generalist scan. High-confidence issues surface clearly; uncertain observations are filtered out. It’s a different category from “ask Claude to review my PR”: the structured multi-agent approach catches the things a single-pass review tends to miss. Confidence thresholds are the same lever we use for deterministic confidence gates elsewhere in the stack.

Particularly useful before merging anything that touches critical paths.

6. Context7: live docs in context

268k installs · claude.com/plugins/context7

Claude’s knowledge has a training cutoff. That’s fine for stable APIs, but not for frameworks that ship breaking changes every few months. Without Context7, asking Claude about the latest Next.js 16 routing API might get you confident, outdated v13 syntax. That confident wrongness is textbook hallucination, and it is expensive precisely because it looks right.

Context7 pulls live, version-specific documentation from source repos and injects it directly into the model’s context at query time. You specify the library and version; it fetches the real docs. The difference in output quality for fast-moving ecosystems is significant. Claude stops hallucinating deprecated APIs and starts referencing the actual current behavior.

If you work with any framework that moves fast (Next.js, LangChain, Tailwind v4, React 19), this one pays for itself in the first session.

7. GitHub or GitLab: close the shipping loop

GitHub: 200k installs · claude.com/plugins/github

GitLab: 26k installs · claude.com/plugins/gitlab

Pick the one that matches your stack, because the idea is the same. Both plugins bring full repo management into Claude Code: create issues, open and review pull or merge requests, search repositories, manage labels, and access CI/CD pipelines, all without leaving the terminal.

The practical value is eliminating context switches. When Claude identifies a bug worth tracking, you create the issue immediately. When the branch is ready, the PR or MR gets opened from the same conversation where the code was written. GitLab users get additional depth on CI/CD pipeline management and wiki access, which makes it particularly useful for teams running their entire DevOps lifecycle through GitLab.

The workflow tightens from “write code → switch to GitHub/GitLab → create PR → paste description” to a single, uninterrupted session. It’s the plugin that makes everything else feel complete.

How to install

Getting any of these takes about 30 seconds:

1. Open Claude Code (run claude in your terminal, or use the desktop app)
2. Type /plugin to open the plugin browser
3. Search for the plugin by name and install

All seven are available in the official marketplace. The full list of 140+ plugins is worth browsing if you use specific platforms like Supabase, Vercel, or Linear, which have dedicated plugins worth adding on top of these. Anthropic also documents how to create and distribute your own marketplace if your team wants a private one.

The takeaway

These seven cover the core of a software development workflow:

• Skill Creator encodes your team’s workflows as repeatable slash commands
• Frontend Design and Playwright build UI and verify it, all in one session
• CLAUDE.md Management keeps project context accurate as the codebase evolves
• Code Review gives structured, confidence-filtered feedback before merging
• Context7 stops the outdated API answers from fast-moving frameworks
• GitHub or GitLab ships without context switching

Start with these. Add vertical integrations (database, deployment platform, monitoring) once the baseline is solid. The compounding effect of the right seven plugins is genuinely different from running Claude Code without them, and it stacks with the other lever we lean on, cutting the tokens each session burns.

Sources

• Claude Code plugins marketplace, the official directory and the source of every install count above
• Claude Code documentation, Anthropic’s own overview of the tool and its plugin system
• Create and distribute a plugin marketplace, Anthropic’s guide to running a private marketplace
• Frontend Design, the most-downloaded plugin in the marketplace at 564k installs
• Context7, the live-documentation plugin at 268k installs