Why Plugins Matter More Than You Think
If you've used Claude Code even a little, you've run into the same ceiling: it's an incredible raw tool that gets 20% better the moment you give it the right scaffolding. Plugins are how the community has been filling that gap — opinionated toolkits that bolt on methodology, memory, review workflows, or domain knowledge that Claude Code doesn't ship with out of the box.
The plugin marketplace grew from ~20 plugins at launch to 101 official plugins by March 2026 (33 built by Anthropic directly). Most are forgettable. A small set have become near-mandatory for professional use.
This is the short list of the six I'd actually install today. In the order I'd install them.
1. gstack (Garry Tan's Setup) — If You Work Alone
GitHub: garrytan/gstack · Stars: ~66,000 · Install time: 30 seconds
gstack is the open-source skill pack that Y Combinator's Garry Tan uses for his own coding. It turns one Claude Code instance into a virtual engineering team: a CEO who rethinks product choices, an eng manager who locks architecture, a designer who catches AI-generic output, a QA lead that opens a real browser, a security officer who runs OWASP + STRIDE audits, and a release engineer who ships the PR.
28 slash commands, each activating a specialized "cognitive mode." You type /review and Claude stops being a general coder and becomes a production-minded reviewer. You type /ship and it stops being a reviewer and becomes a release engineer.
The thing that's controversial about gstack — and the reason it has both 66K stars and a vocal hater contingent — is how opinionated it is. It enforces TDD, forces you to answer "what's the customer value?" before writing code, and refuses to let you skip a review step. People who already work that way love it. People who just want to ship a prototype fast find it meddling.
Install if: you work solo or in a 2-3 person team and you want the discipline of a larger organization without the headcount.
2. Superpowers (obra) — If You Want TDD Non-Negotiably
GitHub: obra/superpowers · Stars: ~42,000 · Install time: 2 minutes
Superpowers is Jesse Vincent's agentic methodology for Claude Code. It's older than gstack (released Oct 2025), more mature, and has spawned a marketplace, a Chrome-control plugin, and an experimental lab.
The philosophy is narrower than gstack but deeper in what it enforces: true red/green TDD, YAGNI (You Aren't Gonna Need It), and DRY. When you ask Superpowers-equipped Claude to build something, it refuses to jump straight to code. Instead it pulls a spec out of you in digestible chunks, shows you an implementation plan a junior could follow, and writes the failing test first.
Accepted into Anthropic's official marketplace in January 2026, so you install via the plugin system — not a git clone into ~/.claude/.
Install if: you find yourself asking Claude for things and getting back code that "works on the happy path" but misses edge cases. Superpowers makes that pattern impossible.
3. claude-mem — Persistent Memory Across Sessions
GitHub: thedotmack/claude-mem · Stars: ~46,000 · Install: npx claude-mem install
Claude Code's biggest daily annoyance: every new session starts from zero. You re-explain your project structure, re-explain your conventions, re-explain why you chose X over Y. claude-mem fixes this by hooking into Claude Code's session lifecycle, capturing what Claude does, and injecting relevant context into future sessions.
Technically: 5 lifecycle hooks (SessionStart, UserPromptSubmit, PostToolUse, Stop, SessionEnd), local SQLite + Chroma vector search, and a 3-layer MCP tool workflow (search, timeline, get_observations) that claims ~10x token savings by filtering before fetching.
Pairs exceptionally well with either gstack or Superpowers — those give you the methodology, claude-mem gives you the working memory of the human who would otherwise enforce it.
Install if: you keep re-typing the same context at the start of every session, or you're burning through your Claude Code session windows too fast.
4. Anthropic's Frontend Design Plugin — If You Ship UI
Source: claude.com/plugins/frontend-design
The official Anthropic plugin for frontend work. Auto-invokes when you ask Claude to build UI, steering the output toward bold design choices, distinctive typography, and real animation details — specifically to avoid the generic AI aesthetic (centered hero, gradient button, muted palette, same 3 fonts every time) that every LLM drifts toward.
If you've ever looked at Claude-generated frontends and thought "this looks like every other AI-built site," that's the problem this plugin is trying to fix. Result is usually closer to a real designer's output — with opinionated spacing, intentional asymmetry, and visible design decisions rather than safe defaults.
Install if: the frontend your AI produces looks generic and you're not sure why.
5. Anthropic's Code Review Plugin — For Every PR
Source: claude.com/plugins/code-review
Another Anthropic-official plugin. Runs structured PR review with specialized sub-agents focused on tests, error handling, type design, code quality, and simplification. Each comment carries a confidence score so you can filter the noise — high-confidence issues surface first, low-confidence flags sit in an expandable section.
The less glamorous but surprisingly useful piece: it catches the stuff you usually catch in self-review but wouldn't because it's 11pm and you want to ship. Second pair of eyes that never gets tired.
Install if: you merge your own PRs without formal review.
6. Security Guidance Plugin — Especially If You're Shipping to Users
Source: claude.com/plugins/security-guidance
The Anthropic security plugin applies OWASP guidelines and secure coding best practices during development. Catches the obvious foot-guns (SQL injection, XSS, weak crypto choices, secrets in logs) before they make it into a commit.
This one genuinely reduces your regret rate. The cost of catching an SSRF bug in commit-time review is zero; the cost of catching it post-deploy is a bug bounty payout and a weekend of refactoring.
Install if: you're writing code that handles user data, auth, or payments.
The Install Order I Recommend
- claude-mem first — persistent memory pays dividends on every other plugin you add
- gstack OR Superpowers (pick one — they overlap) — methodology layer
- frontend-design — if you ship UI at all
- code-review + security-guidance — before you push a PR that goes to production
Installing all at once is fine technically. But you get more value by living with each for a few days and learning what it does before adding the next — otherwise you're just drowning in slash commands you don't remember configuring.
Why This Matters for the Cert / Career Side
If you're prepping for the Claude Certified Architect exam, Domain 3 ("Claude Code Configuration & Workflows") is 20% of the exam. Every plugin above teaches you something that shows up on that exam — plugin installation, hooks, skill files, slash command design, MCP integration. Installing and using two or three of them is genuinely good prep.
Free Resources That Pair With Plugins
- Claude skills library (2,300+ skills) — skills are the unit that most plugins compose into. Understanding skill file format makes plugin internals legible.
- Free 40-page Claude guide — covers Claude Code setup, skill files, MCP, hooks. Read this first if plugin internals feel opaque.
- Graphify integration — not a plugin, but a complementary tool. Plugins change how Claude works; Graphify changes what Claude sees.
FAQ
How many Claude Code plugins should I have installed at once?
Three to five is the sweet spot for most people. More than that and you stop remembering what each one does.
Do plugins slow Claude Code down?
Minimally. The biggest perf hit is actually from skill files being re-parsed on session start, not from plugin code. claude-mem adds the most overhead (it's doing real work) but the 10x token savings usually more than pay for it.
Can I write my own plugin?
Yes. The plugin format is documented in anthropics/claude-code. A working plugin is essentially: a .claude-plugin/marketplace.json + a set of skill files + optional hooks + optional slash commands. Many plugins in the marketplace are ~200 lines total.
Are plugins the same as skills?
No. Skills are individual markdown files with specific domain knowledge. Plugins bundle skills + commands + hooks + sometimes MCP servers into an installable unit. A plugin typically ships 5-30 skills; a skill is a single file.
Which one of these 6 plugins should I try first?
claude-mem. It provides value immediately regardless of your workflow, and you'll feel the improvement within one session. Everything else requires you to adapt to the plugin; claude-mem adapts to you.
The Short Version
The Claude Code plugin scene went from hobby project to real ecosystem in about six months. If you're writing code with Claude every day and you haven't installed any plugins, you're using a tool designed for 2025 in 2026. Start with claude-mem for memory, add gstack or Superpowers for methodology, and layer in Anthropic's official frontend-design / code-review / security-guidance as your work demands them.