12 Best Claude Code Plugins in 2026 (Marketplaces, Install Commands, What to Skip)

By Rakshit Yadav (@yadavrakshit60)•Updated Nov 2026•9 min read

Short answer: Claude Code plugins (installed via /plugin marketplace add + /plugin install) bundle agents, skills, commands, and hooks into one shareable package — they're the newest distribution surface in Claude Code, and most "best plugins" lists ranking for this right now are thin affiliate posts that never actually ran /plugin install on anything. Below are 12 picks across workflow automation, testing, docs, and security, split by official vs. community marketplace, plus the audit steps to run before you install anything that can execute code in your repo.

Plugins vs. skills vs. MCP vs. agents — what a plugin actually bundles

These four pieces of Claude Code get conflated constantly, and the distinction matters for what you're trusting when you install one:

  • A skill is an instruction file (SKILL.md) the agent chooses to follow when its description matches the task — no new capability, just discipline for tools Claude Code already has.
  • An MCP server gives the agent new tools — live access to a real API or database (Stripe, Postgres, GitHub) — connected with claude mcp add. See the MCP servers guide for the setup commands and the audit checklist that also applies here, or the curated, not-context-hungry shortlist if you want the context-cost filter for deciding which servers to connect in the first place.
  • A subagent is a dedicated persona with its own system prompt and tool scope, invoked via the Agent tool for a specific kind of work (code review, test writing).
  • A plugin is the distribution wrapper around all three — a single /plugin install can drop in commands, skills, subagent definitions, and hook scripts at once, sourced from a marketplace (a git repo or URL listing available plugins).

That last part is why plugins carry more risk than a single skill file: installing one can register hook scripts that run on every tool call, not just inert prompt text. Read what's in the bundle before you install it, the same way you'd read an MCP server's source before connecting it.

What is a Claude Code plugin, singular?

A Claude Code plugin is one installable package — one /plugin install <name>@<marketplace> — that can bundle a command, a skill, a subagent definition, and a hook script together under a single name. People search "claude code plugin" (singular) when they mean one specific package, and "claude code plugins" (plural) when they mean the ecosystem or a list like this one. The confusion that actually trips founders up isn't singular vs. plural, it's plugin vs. skill: a skill is just a markdown playbook the agent opts into; a single plugin can ship a skill plus a hook that runs unconditionally. Installing "a plugin" is a bigger trust decision than installing "a skill," even when the plugin's README reads the same.

How to add a marketplace and audit a plugin before installing

# Add a marketplace (official Anthropic one, or any git repo/URL following the marketplace schema)
/plugin marketplace add anthropics/claude-code

# List what's available
/plugin marketplace list

# Install a specific plugin
/plugin install <plugin-name>@<marketplace-name>

# See what's currently installed
/plugin list

# Remove one
/plugin uninstall <plugin-name>

Before you run /plugin install on anything from a marketplace you don't already trust:

  1. Read the marketplace repo's marketplace.json and the plugin's source, not just its README. A plugin is a folder of .md files, commands, and optionally hook scripts — it's usually small enough to read in five minutes.
  2. Check for hook scripts specifically. A skill or command is inert until the agent chooses to use it; a hook can execute on every PreToolUse or PostToolUse event, which is a different trust level entirely.
  3. Check commit history and maintenance, not star count. A marketplace repo with no commits in six months and no changelog is a plugin you're installing blind.
  4. Prefer the official anthropics/claude-code marketplace or a maintainer you already recognize for anything that touches credentials, deploys, or git history.
  5. Uninstall what you're not using. Same discipline as MCP servers — /plugin list periodically and remove anything installed to test once and forgotten.

12 plugins worth installing in 2026

#PluginMarketplaceCategoryWhat it does
1commit-helperanthropics/claude-code (official)WorkflowStructured commit-message generation following conventional-commits, wired to a pre-commit hook
2pr-revieweranthropics/claude-code (official)WorkflowAdds a /review-pr command that diffs against the base branch and reports findings inline
3test-writercommunityTestingGenerates test scaffolding from a changed file's exports; framework-aware (Vitest/Jest/pytest)
4coverage-gatecommunityTestingHook-based: blocks a commit if changed lines drop test coverage below a threshold you set
5docs-synccommunityDocsFlags functions/endpoints with stale or missing docstrings after a diff, via a PostToolUse hook
6changelog-writercommunityDocsBuilds a changelog entry from merged PR titles; pairs well with a release command
7secret-scannercommunitySecurityHook that greps staged files for common credential patterns before a commit lands
8dependency-auditcommunitySecurityWraps npm audit / pip-audit output into a prioritized, agent-readable summary
9stripe-opscommunitySaaS-specificCommands for common billing support tasks (look up subscription, check webhook delivery) — audit this one closely since it touches money
10migration-assistantcommunitySaaS-specificGenerates and reviews database migrations against your existing schema conventions
11onboarding-packcommunityWorkflowBundles a new-contributor walkthrough command plus a repo-structure explainer skill
12incident-respondercommunitySREAdds a /triage command that pulls recent error logs and drafts a root-cause timeline

Anthropic's official marketplace (anthropics/claude-code) carries the most scrutiny since it ships from the vendor itself — start there for #1 and #2. Everything else above is a community plugin: audit the repo per the checklist above before installing, especially #7–#10, which touch credentials, dependencies, or billing.

Governance and approval gateway plugins

If you're installing plugins for a team rather than a solo repo, the category to look for is a governance and approval gateway: not a single named plugin, but a managed-settings.json allowlist that restricts which marketplaces and plugins a team is permitted to install from, enforced centrally rather than left to whoever typed the install command. See the plugin marketplace governance guide for how to set that allowlist up and when a solo founder can skip it entirely — usually until you have a second person with their own Claude Code seat installing plugins into the same repo.

Red flags to skip

  • No changelog, no commit history past the initial push. A plugin that hasn't been touched since launch is one nobody is maintaining when Claude Code's plugin schema changes.
  • Auto-generated "awesome-claude-code-plugins" lists with 80+ entries and no usage notes. These are typically scraped GitHub topic searches, not a maintainer's tested picks — the volume is the tell.
  • A plugin that bundles a hook script you can't read in under five minutes. If the hook logic is obfuscated, minified, or calls out to an external endpoint, that's a plugin to skip regardless of what it claims to do.
  • Marketplaces with no marketplace.json schema validation or mismatched plugin names. A sloppy manifest is often a sign the rest of the repo wasn't reviewed either.
  • Review sites ranking plugins they clearly never installed. If a "best plugins" post can't show the actual /plugin install command or describe what changed in /plugin list after, it's filler — several of the current top results for this query are exactly that.

There's no binary signing for Claude plugins yet, which is exactly why the checklist above matters more than star count — see the governance breakdown for the supply-chain risk and when a managed-settings allowlist is actually worth setting up.

Where AgenticKit's 34-agent kit fits vs. installing plugins piecemeal

Plugins are the right tool when you want one specific capability — a commit hook, a changelog generator — without touching anything else in your setup. AgenticKit is a different scope: one install drops in 34 subagents, 61 skills, and 47 commands designed to work together (a tech-lead agent that hands off to test-automator, a /ship command that chains the skills a release actually needs), rather than a single bundled feature. If you're assembling a plugin-by-plugin stack for commits, tests, and docs separately, the full agent list and skill list cover the same ground pre-wired, with the review discipline from code-reviewer and security-auditor already built in rather than something you have to vet plugin-by-plugin.

FAQ

What's the difference between a "claude code plugin" and a "claude code skill"? A skill (singular file, SKILL.md) is inert until the agent's task matches its description — no new risk beyond the instructions it contains. A plugin is the installer wrapper: one /plugin install can register a skill, a command, a subagent, and a hook script together, and that hook can run on every matching tool call regardless of whether the agent "chose" to use it. If a listing just says "plugin," check whether it ships a hook before treating it like a skill-level install.

Are Claude Code plugins the same as MCP servers? No. An MCP server gives the agent new tools for live external access (a real database, a real API). A plugin is a bundle of commands, skills, subagents, and hooks — it can register an MCP server as part of its install, but most plugins don't touch external systems at all.

Where do official Anthropic plugins live? In the anthropics/claude-code marketplace, added with /plugin marketplace add anthropics/claude-code. That's the marketplace to trust by default for anything touching your git history or commit flow.

Can a plugin see my code or send it anywhere? A plugin's commands and skills run as instructions the agent follows with tools it already has access to — same permission boundary as a slash command you wrote yourself. A hook script, though, runs arbitrary code on tool-call events, so that's the part worth reading before installing, not the markdown prompt files.

Do I need plugins if I already use AgenticKit? Not for the overlapping ground — AgenticKit's agents, skills, and commands already cover commit discipline, test writing, and docs the way the plugins above do. A plugin still makes sense for something narrow and Claude Code-native that AgenticKit doesn't model as an agent, like a specific hook-based gate.

How do I know if a plugin is still maintained? Check the marketplace repo's commit history and whether marketplace.json has been updated for the current plugin schema — a repo with no commits in 6+ months and no changelog is the main red flag listed above.

Next steps

Start with the official anthropics/claude-code marketplace for commit and PR-review plugins, since those carry vendor-level trust. For testing, docs, and security coverage beyond that, either audit a community plugin against the checklist above, or skip the piecemeal approach and pull in AgenticKit's agents and skills, which cover the same categories pre-reviewed. If you're also connecting live tools rather than bundled instructions, the MCP servers guide covers that adjacent surface and the same supply-chain checklist applies.

Related Tools & Agents

🛠️ Free Tool: kit-picker🤖 Agent: code-reviewer🤖 Agent: security-auditor🤖 Agent: test-automator⚡ Command: /ship⚡ Command: /auditSkill: security-basicsSkill: playwright-e2eSkill: technical-docs

Skip plugin-by-plugin assembly

AgenticKit ships 34 agents, 61 skills, and 47 commands pre-wired to work together, covering the same ground a plugin stack would piece together one install at a time.

Browse agents →