Nine thousand Claude Code plugins. That is how many showed up on the community side of the marketplace by the time I sat down to write this. I stared at the list the way I used to stare at the back wall of a record shop in 2008, flipping through crates of sleeves I did not recognize, pretending I knew what I was looking for.
I did not know what I was looking for.
I had just run /plugin marketplace browse for the third time that morning, and the results scrolled past like a ticker tape written by someone fluent in a language I half-remembered. Linters. Subagents. Hooks. Skills that invoke other skills. A plugin that promised to "rewrite your prompts in the voice of a medieval scribe," which, reader, I installed immediately and uninstalled twenty minutes later.
The record shop comparison is not accidental. App stores feel too clean. Plugins, even the official ones, have the same slightly-loved, slightly-weird energy as used vinyl: someone made this for themselves first, uploaded it second, and hoped you might like it too. Some of it is brilliant. Some of it is a B-side that should have stayed on the cutting room floor. The trick, as always, is knowing which is which before you have spent forty minutes configuring a hook that mostly prints ASCII cats.
So this piece is my attempt to save you from the crate-digging I already did.
What plugins actually are, and why you probably want some
Claude Code plugins are the thing I kept hearing about at meetups last fall and kept putting off learning. They bundle four pieces together: skills (reusable prompts with file references), subagents (scoped Claude instances), slash commands, and hooks that fire on events like PreToolUse or Stop. You install one, and the whole bundle becomes part of your session like it was always there.
Anthropic launched the plugin ecosystem in public beta back in October 2025, and by the time the official marketplace opened its doors properly on February 24 this year, there were already 101 officially curated plugins and, depending on how you count forks and duplicates, somewhere north of nine thousand in the community index. Thirty-three of those officially curated plugins are built by Anthropic themselves, spanning language servers, dev workflow, and setup tools. Enterprises get private marketplaces. Indie devs get the wild open one.
Installing is simple. From inside a Claude Code session you just type /plugin install @marketplace/plugin-name, or you can browse with /plugin and hit enter on whichever one catches your eye. If you want the deeper setup, the plugins repo on GitHub has the marketplace manifest format and examples. Private marketplaces work the same way, just scoped to a URL your IT team controls, which is how most enterprises are rolling this out.
If you are brand new to Claude Code itself, start with the beginners tutorial and come back. If you are past the tutorial and want to keep going deeper, the 50 Claude Code tips piece is probably the best companion to this one.
A quick word on how I picked these. I have a notes file called plugins-to-try.md that has been growing since October. Every time somebody mentioned a plugin on Discord, or I saw one in a PR description, or somebody retweeted a screenshot, it went on the list. Over about four months I installed roughly sixty of them. Most lasted a week. A handful lasted a day. These twenty lasted.
The twenty below are the ones I actually use. Not the ones I installed and forgot about. Not the ones with the prettiest READMEs. The ones I would be annoyed to lose tomorrow.
Productivity and workflow
These are the claude code plugins I boot up on every new project, no exceptions.
@anthropic/devflow turns a messy repo into a structured workflow. It reads your README, your open PRs, and your issue tracker, then proposes a weekly plan with /plan-week and tracks it across sessions. The subagent piece spins up a lightweight Claude that only touches planning docs, so your main session stays focused on code.
I use this every Monday morning. It replaced about three Notion pages and a Calendly reminder I had set for myself.
- Install:
/plugin install @anthropic/devflow - Verdict: Essential. If I had to keep only one plugin, this would be it. My one complaint is that it occasionally invents issue numbers that do not exist, so I still sanity-check its output.
@workshop/commit-crafter is a commit message plugin that actually gets the hang of your team's style after about a dozen commits. It hooks into PreToolUse on git operations and proposes conventional commit messages with real context, not the "update file.ts" slop I was getting before.
- Install:
/plugin install @workshop/commit-crafter - Verdict: Solid but opinionated. If you hate conventional commits, skip it.
@anthropic/session-memory is a long-lived memory layer that persists notes across sessions. You tell Claude something once ("we use Vitest, not Jest"), and it remembers in week three without needing to re-read the whole repo.
- Install:
/plugin install @anthropic/session-memory - Verdict: Useful but noisy. I had to tune which files it indexes, otherwise it kept pulling stale notes from an old client project. Worth the tuning.
@community/pomodoro-pair is the silly one I kept. It breaks work into 25-minute focus blocks, uses hooks to stop Claude from starting long-running tasks near the end of a block, and drops a little ASCII coffee cup in the terminal when it is time to rest. Simple hooks, real behavior change.
- Install:
/plugin install @community/pomodoro-pair - Verdict: Gimmicky on paper, genuinely helpful in practice. I get more done when my pair programmer knows when to shut up. I turned off the coffee cup. I left everything else.
Language support
The language server wrappers are where the ecosystem really started to feel mature to me. Before plugins, I was copying the same "install pyright and configure it" instructions into every Python project. Now it is one line. The irony of having an AI pair programmer that did not know your types was, frankly, the thing that nearly put me off the whole category last year. Plugins fixed it.
@anthropic/python-lsp bundles pyright, ruff, and a Claude-aware type refactoring subagent. When Claude writes Python, the plugin catches type errors before the file is even saved, and the subagent proposes fixes that respect your existing patterns.
- Install:
/plugin install @anthropic/python-lsp - Verdict: The Python experience I wish I had in 2023. The subagent occasionally over-engineers generic types, but that is a taste issue.
@anthropic/ts-toolkit does the same for TypeScript, wrapping tsserver, Biome, and a repo-aware refactor subagent. It understands your tsconfig.json paths and does not hallucinate imports from node_modules that do not exist.
- Install:
/plugin install @anthropic/ts-toolkit - Verdict: I stopped manually fixing import paths about a week after installing this. Huge quality of life jump.
@anthropic/go-companion integrates gopls, staticcheck, and a subagent that specifically handles the "Claude wrote idiomatic Python that happens to compile as Go" problem I run into every time I switch languages mid-project. It nudges toward Go idioms: pointer receivers, error returns, small interfaces.
- Install:
/plugin install @anthropic/go-companion - Verdict: Does exactly what it says. If you write Go more than twice a month, install it.
@community/rust-analyzer-plus wraps rust-analyzer and adds a borrow-checker explainer subagent. I am not going to pretend I understand everything it tells me, but when Claude proposes a change that would not compile, this plugin catches it before I waste tokens arguing with it. The explainer subagent is what sold me: it translates borrow-checker errors into short paragraphs about what the compiler is protecting you from, not just what rule you broke.
- Install:
/plugin install @community/rust-analyzer-plus - Verdict: Honestly not sure if it will still work in six months, because the maintainer is one person with a real job and an occasionally grumpy changelog. But right now, it is the best Rust support I have used.
Integrations
These connect Claude Code to the other tools you probably already live in. Your mileage here varies more than in the other categories, because it depends entirely on which tools your team already pays for. I use Linear and GitHub daily. I use Jira only when a client makes me. I use Slack more than I would like to.
@anthropic/github-ops is the official GitHub integration, and it replaced the scrappy gh CLI wrappers I had been chaining together. It handles PR reviews, issue triage, draft PR creation, and, the feature I did not know I wanted, a /gh-context command that pulls the last week of repo activity into the session so Claude actually knows what has been happening.
- Install:
/plugin install @anthropic/github-ops - Verdict: Pair this with the beginners tutorial if you are just getting started with Claude Code and git at the same time. The PR review subagent alone is worth the install.
@linear/claude-sync is Linear's first-party plugin. It syncs your assigned issues, lets you move them across states from inside Claude Code, and, critically, embeds the Linear issue context into whatever code session you open for that ticket. Branch names auto-generate from the issue ID.
- Install:
/plugin install @linear/claude-sync - Verdict: If you use Linear, this is a no-brainer. If you use Jira, keep reading.
@atlassian/jira-bridge is Atlassian's equivalent, and it is fine. It does what Linear's plugin does but slower, and the JQL query builder is still clunky. I keep it installed because my day job uses Jira, and it beats the browser.
- Install:
/plugin install @atlassian/jira-bridge - Verdict: Better than context-switching to the Jira tab. Lower than Linear's bar. But Atlassian is shipping fast, so this may age well.
@slack/claude-chatops lets you kick off Claude Code sessions from Slack and get notifications back. I use it to trigger a "run the full test suite and report" agent when I merge to main, and it posts the summary in the deploys channel. Saves me from babysitting CI tabs. The slash command surface is small on purpose, which I appreciate. Nobody wants to learn another DSL just to ask if the build is green.
- Install:
/plugin install @slack/claude-chatops - Verdict: Surprisingly reliable. The permission model made me nervous at first, so I read the security docs twice before enabling it on our production repo. Worth that ten minutes.
Security and code review
This is the category where I was most skeptical and most converted. Automated security tools have a long history of crying wolf, and I assumed plugin versions would be worse. Some are. These four are not. They share a common trait: they interrupt only when they should, and they tell you exactly why when they do.
@anthropic/secret-scout scans your diffs before commit for leaked API keys, tokens, and credentials. It uses a local hook on PreToolUse for git, so the check happens before the commit lands, not after. I have had it catch two real keys in the three months I have used it, both of mine, both embarrassing.
- Install:
/plugin install @anthropic/secret-scout - Verdict: Install this today. Not tomorrow. Today. The false positive rate is low enough that I do not tune it out.
@snyk/claude-audit brings Snyk's vulnerability database into the session. When Claude proposes a new dependency, the plugin checks it against known CVEs and flags anything sketchy before the npm install runs.
- Install:
/plugin install @snyk/claude-audit - Verdict: The free tier is enough for solo devs. Teams will hit the paid plan fast, but that is true of any security tooling.
@community/code-reviewer is a community-built review subagent that reads a PR diff and comments like a tired but fair senior engineer. It is not as thorough as a human review, but it catches the obvious stuff: missing error handling, duplicated logic, and inconsistent naming. Think of it as the friend who reads your cover letter before you send it, not the hiring manager.
- Install:
/plugin install @community/code-reviewer - Verdict: I use it on my own branches before I open a PR. It saves me from the "oh no, everyone can see that typo" moment. That alone is worth the install.
@anthropic/policy-check lets you encode team coding policies ("no any in TypeScript," "all database calls through the repository layer") as checkable rules. The plugin runs them as hooks and warns before Claude writes code that violates them.
- Install:
/plugin install @anthropic/policy-check - Verdict: Enterprise-flavored but genuinely useful for solo devs too. I use it to enforce my own past decisions against my own future lazy self.
Formatting and quality
The last category is the least glamorous and the one I miss most when it is gone. Formatting is a little like the bass player in a band, you notice its absence more than its presence, and a good one holds the whole groove together without drawing attention to itself.
@biomejs/biome-plugin runs Biome on every file Claude writes, instantly. No config, no bikeshedding. The formatting happens in a PostToolUse hook so the session never waits for it.
- Install:
/plugin install @biomejs/biome-plugin - Verdict: If you have not tried Biome yet, this is a good excuse. It is fast in a way that makes Prettier feel like it is wading through mud.
@anthropic/test-scaffold generates test files alongside new code, using whatever framework it detects in your repo (Vitest, Pytest, Go's testing package, and so on). It does not write the tests, exactly, it writes the scaffolding and some obvious cases, and leaves the interesting ones for you.
- Install:
/plugin install @anthropic/test-scaffold - Verdict: The honest framing of "I wrote the boring tests, you write the real ones" is refreshing. I was worried it would try to do too much.
@community/docstring-doctor scans for missing or stale docstrings and proposes updates. It uses the signature and surrounding code to draft something reasonable, then asks before committing.
- Install:
/plugin install @community/docstring-doctor - Verdict: Good for catching drift on mature codebases. Less useful on greenfield work, where Claude is already writing decent docstrings as it goes.
@anthropic/diff-guardian is the plugin I almost left off this list because it sounds boring and is not. It previews every non-trivial diff in a readable format before the change lands, with a short natural-language summary of what changed and why. I catch roughly one "wait, that is not what I asked for" moment per day because of this. At fifteen minutes of cleanup saved per catch, that is nearly two hours a week I am no longer un-doing work Claude did while I was looking the other way.
- Install:
/plugin install @anthropic/diff-guardian - Verdict: Makes Claude feel less like a coworker who commits while I am on lunch. It waits for my nod, and I am more trusting as a result. The Anthropic engineering blog has a piece about the design philosophy behind this one, and it is worth the five minutes if you want to understand why the interruption points are where they are.
What I am still figuring out
Here is where I admit the thing I keep circling. The plugin ecosystem is moving fast enough that I do not fully trust this list to be correct in six months. Maintainers get bored. Anthropic ships better first-party versions of community plugins. Last week I found out my favorite markdown linter plugin had been deprecated in favor of a new one I had never heard of. The crate-digging never stops.
That is also the fun of it. The record shop was fun because you did not know what you would walk out with. The Claude Code plugin marketplace has the same energy, if you squint. You will find a plugin that solves a problem you did not know you had. You will install a plugin that claims to do something magical, discover it is actually just a sed command in a trenchcoat, and move on. And every so often, you will install one and realize, an hour later, that you have quietly rearranged how you work.
I do not know if the plugin model will look the same a year from now. I suspect it will not. I suspect the really good claude code plugins will consolidate, the weird ones will proliferate, and somebody will build a plugin whose only job is to manage your other plugins, and we will have invented package managers all over again. There is a joke in there about the Unix philosophy finally arriving at the agent layer, but I am going to let somebody else make it.
A thing I want to say to teams considering the private marketplace route: it is worth it. The first private marketplace I set up for a client took about forty-five minutes, most of which was arguing about naming conventions with myself. Once it was up, I could ship a plugin that encoded that team's coding policies, their deploy conventions, and their incident runbook, and onboard a new engineer in the time it takes to run /plugin install. The plugin model is, among other things, a very good documentation delivery mechanism for the kind of tacit knowledge that never quite makes it into the README.
But that is future-us problem. Present-us has twenty install commands and an afternoon.
If you run any of these and hit something I missed, or if you have a plugin I should have on the list and do not, tell me. I am still crate-digging. I suspect the list I write in six months will be half-overlapping with this one, and I am curious which half. The stack is never quite finished, which is, mostly, the point. Try a few. Tell me what you find.
