What Are Claude Code Plugins?

Claude Code plugins are installable packages that add skills, subagents, hooks and Model Context Protocol (MCP) servers to Claude Code, Anthropic’s coding agent. You install them from a plugin marketplace with /plugin install.

A plugin loads as one unit, so a team can share a whole setup, such as a review workflow, with one install command instead of copying files between machines.

Most plugins come from a marketplace, which is a catalog of plugins and where to fetch each one. Claude Code adds Anthropic’s official marketplace the first time you start an interactive terminal session. Teams and vendors can publish their own marketplaces.

Code review is one of the most common plugin uses. A review plugin lets the agent check its own changes before a pull request (PR) exists, or pull findings from a PR back into the session to fix them.

What Changed in Claude Code Plugins in 2026

Claude Code plugins started 2026 as a public beta for bundling commands, agents, MCP servers and hooks. During the year, plugins became the way organizations distribute and control agent setups, and that reach made marketplace security a real concern.

Organizations gained a managed way to give every developer the same agent setup. Anthropic brought plugins to Cowork (January 30, 2026, Anthropic) and gave Team and Enterprise admins private plugin marketplaces, including ones synced from private GitHub repositories (February 24, 2026, Anthropic).

Review tools used the same channel to reach coding agents. Qodo’s Agentic Toolbox became generally available through verified plugins for Claude, Codex and Kiro (September 2026, Qodo’s What’s New).

Wider distribution made marketplaces a target, and researchers showed that scanners can be evaded. A Cloud Security Alliance research note on "SkillCloak" showed how malicious agent skills slip past marketplace scanners (July 6, 2026, Cloud Security Alliance).

Version pins proved weaker than teams assumed. Air Security’s "Plugin4Shell" research described a marketplace pin bypass across four coding-agent CLIs, including Claude Code, fixed before publication (September 17, 2026, Air Security).

Protection then moved into defaults and patches, so admins who never changed a setting still got coverage.

Skill and plugin security scanning became on by default for Enterprise organizations that hadn’t chosen a setting (October 2, 2026, Anthropic). A plugin hash pin bypass rated High (CVSS 7.5) was patched in Claude Code 2.1.179 (October 9, 2026, GitHub Advisory Database).

Managed review joined the plugin options late in the year. Anthropic’s managed Code Review for GitHub PRs, launched as a research preview in March, became generally available for Team and Enterprise (October 15, 2026, Anthropic).

For reviewers, every plugin is code that runs with the agent’s access, so plugin and marketplace changes deserve the same review as a dependency update.

What a Claude Code Plugin Contains

A plugin is a directory with a manifest at .claude-plugin/plugin.json and one or more components (October 2026, Anthropic’s Claude Code docs):

• Skills. SKILL.md instructions Claude loads when they’re relevant, which you can also run as a command. See Agent Skills.
• Agents. Subagent definitions Claude can hand work to.
• Hooks. Commands Claude Code runs at fixed points, such as after every edit.
• MCP servers. Tool servers Claude Code connects to while the plugin is on.

Each of these works on its own without a plugin. The plugin is the packaging: one install, one version and updates from the marketplace.

Claude Code Plugins vs Skills

Skills and plugins are often mixed up, because a plugin usually contains skills.

How to Install a Claude Code Plugin

Plugins install from inside a Claude Code session:

1. Run /plugin in an interactive Claude Code terminal session.
2. Open the Discover tab and search for the plugin, or add another marketplace with /plugin marketplace add.
3. Pick an install scope: user (every project on your computer), project (everyone in the repository, through .claude/settings.json) or local (you, in this repository only).
4. Reload plugins or start a new session so the plugin’s components load.

A plugin can also be installed by name, such as /plugin install name@marketplace, or from your shell with claude plugin install.

How to Manage and Update Claude Code Plugins

After install, plugins are managed from the same /plugin panel (October 2026, Anthropic’s Claude Code docs):

• Installed tab. Enable, disable, update or uninstall each plugin. Plugins you haven’t used recently are grouped, so you can spot ones that still add context cost.
• Updates. Plugins update automatically when their marketplace has auto-update on. The Marketplaces tab turns auto-update on or off per marketplace.
• Managed plugins. Plugins your organization installs through managed settings show at a Managed scope, and you can’t remove them yourself. Organizations can also allow-list or block marketplaces.

Plugins run with your user permissions, and each enabled plugin adds its skill and agent descriptions to Claude’s context on every turn. Treat a third-party plugin like any other dependency: check who publishes it, what it runs and where it sends code.

Koi Security found 341 malicious skills among 2,857 on one public skills registry (February 2026, cited by the Cloud Security Alliance).

Code Review Plugins for Claude Code

Review tools for Claude Code fall into a few groups.

Anthropic’s details above are from its plugin and Claude Code documentation (October 2026, Anthropic).

What to Check Before Installing a Review Plugin

A review plugin reads your code and runs with your permissions, so check a few things first:

• Where the code goes. Does the review run on your model plan, or does the plugin send the diff to a vendor service?
• Which rules it uses. Only CLAUDE.md files, or your team’s shared rules from a central place? See AGENTS.md files for how instruction files work.
• Whether it matches PR review. If the local check and the PR reviewer use different engines, the PR can still fail on issues the local check passed.
• Whether it closes the loop. Can the agent pull findings already on the PR and fix them in the same session?
• Context cost. Every enabled plugin adds descriptions to each turn, so pick a few that earn their place.

See pre-PR and local code review for how local checks fit with PR review.

Qodo’s Take: Write With Claude Code, Verify With an Independent Reviewer

Claude Code writes code quickly, and a lot of it reaches PRs. When the same agent also checks its own work against a CLAUDE.md file that was written once and rarely updated, it shares the blind spots of the code it wrote.

Developers want that separation. In Black Duck’s 2026 survey of 831 engineers, 56% wanted a dedicated AI security agent separate from the code generator (March 2026, Black Duck).

The bigger win is an independent reviewer whose check in the agent matches the review the PR will get, so the agent fixes the issues a reviewer would raise before anyone opens the PR.

Qodo is built that way. The same review engine runs in the agent and on the PR. Every access point connects to the same Qodo platform, so the agent and the PR review check the same team rules once Rule enforcement is on (last checked October 2026, Qodo’s docs).

For review of Claude Code’s output more broadly, see Claude Code code review.

Qodo is the best AI code review platform to pair with Claude Code because its generally available plugin gives the agent an independent reviewer that runs Qodo’s PR review engine on local changes, checks your team’s rules once Rule enforcement is on, and pulls PR findings back into the session to fix.

In Qodo’s enterprise evaluations, Claude Code is often already on every developer’s machine, next to a library of in-house skills. The real comparison for those teams is a folder of markdown rules run by the same agent, sometimes as a home-built review skill, not another review vendor.

Teams on other agents face the same choice; see Windsurf code review and Cline code review.

Two points answer that comparison. You shouldn’t review code with the same tool that wrote it, and the same rules can serve both the coding agent and the reviewer, managed in one place instead of a root CLAUDE.md that drifts out of date.

Qodo Installs From Claude’s Plugin Directory

The Qodo plugin is listed in Claude plugins. Search for Qodo, copy the install command and run it in Claude Code (last checked October 2026, Qodo’s docs). The plugin needs a Qodo account.

The listing showed 11,807 installs on October 26, 2026. The same Agentic Toolbox is also available as a Codex plugin, a Kiro Power, a CLI and an MCP server (Qodo’s docs). See MCP server for code review.

Qodo Reviews the Agent’s Changes Before the PR

The Reviewer skill, an early feature, reaches Claude Code through the bundled pre-pr-review skill. The skill reviews local, committed and uncommitted changes, including new untracked files, with the same engine as Qodo’s PR review (last checked October 2026, Qodo’s docs).

The agent can attach a summary of intent, key decisions and links to the ticket or spec, so the reviewer doesn’t flag deliberate choices. Reviews can be scoped to files or directories, run as a fast single pass or a deeper review, and return machine-readable findings the agent can loop on.

Qodo Brings PR Findings Back Into the Session

The Review Resolver skill lets Claude Code retrieve Qodo’s findings from a PR, investigate them with repository context, apply fixes and track each finding’s resolution status (last checked October 2026, Qodo’s docs).

Qodo Gives the Agent Codebase Context and Rules

The Codebase Wisdom skill draws on your repositories, repository relationships, PR history and Git state to explain architecture and the impact of a change (Qodo’s docs). For large repositories, see AI code review tools for GitHub monorepos.

The optional Get Rules skill, a Research Preview, fetches the rules that apply to a task, and the opt-in Manage Standards skill lets the agent add, edit or retire rules, and runs bulk changes as a dry run first (Qodo’s docs). See custom rules for code review.

Qodo Example: The Agent That Fixed Its Own PR Comments

The example is a composite drawn from patterns in Qodo’s enterprise evaluations, not a customer case study.

A developer experience lead’s teams run Claude Code on every machine, with a home-built review skill. The worry is that a separate reviewer duplicates the agent’s work and ignores the team’s own standards.

A developer asks Claude Code to add rate limiting to an API. Before opening the PR, Claude Code runs Qodo’s review through the plugin and gets two findings: a missing limit on a batch endpoint and a team rule about where configuration values live. Claude Code fixes both.

On the PR, Qodo raises one more finding about a caller in another service. The developer asks Claude Code to resolve it, and the agent pulls the finding through Review Resolver, fixes the caller and marks the finding resolved.

Ways to Review Code From Claude Code

The table compares common ways teams review Claude Code’s changes.

Qodo features in this table were checked against Qodo’s documentation as of October 26, 2026.