Skip to content

Claude Code plugin install now adds the marketplace in the same command

Update/ 15 min read

Claude Code 2.1.292 installs a plugin from a marketplace you have not added in one command and lets Claude set a subagent's effort per call. Syntax, security, version checks.

  • AI
  • Claude Code
In this article (8)
  1. What changed in Claude Code 2.1.292 (and what did not)
  2. Install a plugin with --marketplace in one command
  3. Security reviews: what 2.1.292 ships and what is still on you
  4. Subagents with an effort level
  5. How to update and check your version
  6. Common mistakes
  7. Frequently asked questions
  8. Checklist

Since Claude Code 2.1.292, released October 6, 2026, a single command installs a plugin from a marketplace you have not added yet: claude plugin install <plugin> --marketplace <source>. The same release lets Claude run a subagent at the effort level you ask for. On security it brings policy checks and fixes, not a new review feature.

Below: what the changelog actually says, the documented syntax for the shell and for a session, what Claude Code checks before it adds a marketplace and what is still on you, how to set a subagent's effort, and how to confirm your version has all of this. We read the changelog and the docs on November 11, 2026, and at the end there is a team marketplace we validated locally.

What changed in Claude Code 2.1.292 (and what did not)

The 2.1.292 entry in the Claude Code changelog is dated October 6, 2026, and the npm registry shows it published that same day. If you saw another date: October 7 was 2.1.293. The entry lists 8 additions, 7 changes, 13 improvements and 64 bug fixes. These are the ones that touch plugins, marketplaces and subagents:

ChangeWhat the changelog saysWhere you notice it
--marketplace <source> on claude plugin installAdds the marketplace if needed, under the same policy checks as claude plugin marketplace add, then installs the plugin from itShell and scripts
effort parameter on the Agent toolClaude runs a subagent at the effort level you ask forInside a session
Policy loads firstclaude plugin commands such as marketplace add and install no longer run before an organization's managed settings load on a first runTeams with plugin policies
Look-alike namesBetter steps when a marketplace declared in a settings file has a name that looks like an official Anthropic marketplaceSettings that declare marketplaces
Agent namesAt most 256 characters; a longer name in a skill or a plugin file is ignoredPlugin authors

Two details before going further. The effort field in a subagent file is not new: 2.1.267, from September 9, already fixed a bug in it. What is new is Claude passing a level on each call. And the session form, /plugin install <plugin> --marketplace <source>, arrived earlier, in 2.1.275 on September 17; 2.1.292 brings the same thing to the shell.

What is not there: the entry adds no feature called "security review". It does ship security fixes, covered below, and the /security-review command is still in the commands list, but the 2.1.292 changelog does not mention it.

Install a plugin with --marketplace in one command

Until now, installing a plugin from a new marketplace took two steps: register the catalog, then install with the marketplace name after the @. The install and manage plugins guide now documents the one-step version. Using the example from the docs themselves:

# Before: two commands
claude plugin marketplace add acme/tools
claude plugin install deploy-helper@acme-tools

# Since 2.1.292: one
claude plugin install deploy-helper --marketplace acme/tools
  1. claude plugin marketplace add acme/tools registers the marketplace from the GitHub repository acme/tools. It registers under the name in its marketplace.json, which in this example is acme-tools, not under the repository name.
  2. claude plugin install deploy-helper@acme-tools installs the plugin. What follows the @ is that marketplace name, and you had to know it in advance.
  3. claude plugin install deploy-helper --marketplace acme/tools does both. The plugin name goes alone, with no @marketplace, and the source takes the same forms as marketplace add: a GitHub owner/repo, a git URL or a local path. If you already added that source, it is reused.

The plugin commands reference requires 2.1.292 or later for the shell flag. The shell and session forms behave differently, and the differences matter:

  • In the shell, the marketplace is added without a confirmation step. When it is new, the command prints Successfully added marketplace: <name> (declared in user settings) and then installs, at user scope unless you pass --scope.
  • The marketplace lands in your user settings even when you pass --scope project. The plugin can be enabled for the whole repository; the marketplace is not shared that way.
  • In a session (2.1.275 or later), /plugin install deploy-helper --marketplace acme/tools shows the source it resolved and asks you to confirm before adding it. Then the plugin's details open so you can pick a scope. There, the source cannot contain spaces.
installing a plugin from a marketplace you have not added yetbefore · two commandsclaude plugin marketplace add acme/toolsregisters the catalog; its name comes from marketplace.jsonclaude plugin install deploy-helper@acme-toolsnow · one command (2.1.292, terminal)claude plugin install deploy-helper --marketplace acme/toolsadds the marketplace if needed, under the same policy checks, then installsin a session (2.1.275+), it shows the source and asks first:/plugin install deploy-helper --marketplace acme/toolsthe shell does not ask to confirm; a session doessource: 2.1.292 changelog (Oct 6, 2026) and the plugin install docs

Scroll the figure sideways to see all of it

A familiar example: with the official Anthropic skills worth installing, the flow was /plugin marketplace add anthropics/skills followed by /plugin install document-skills@anthropic-agent-skills. Following the documented syntax, from the shell it becomes one line: claude plugin install document-skills --marketplace anthropics/skills. We did not run it: our local version predates 2.1.292, and the rule for this test was to install nothing.

When the marketplace belongs to your team

A marketplace is a folder or a repository with .claude-plugin/marketplace.json. The guide to create a marketplace and the marketplace.json reference require three fields, name, owner and plugins, and every plugin entry needs a name and a source. This is the one we built locally, with one plugin in it:

{
  "name": "acme-team-plugins",
  "description": "Internal plugins for the team",
  "owner": {
    "name": "Platform team",
    "email": "platform@example.com"
  },
  "plugins": [
    {
      "name": "reviewer",
      "source": "./plugins/reviewer",
      "description": "Reviewer subagent and PR checklist",
      "category": "review"
    }
  ]
}
  1. name is what people type after the @. It allows letters, digits, ., _ and - only, and it cannot be a reserved name such as claude-plugins-official.
  2. description is optional, but claude plugin validate warns when it is missing.
  3. owner is required, and so is its name; email and url are optional.
  4. plugins holds one entry per plugin. source is a path relative to the marketplace root (the folder that contains .claude-plugin/): it starts with ./ and cannot contain ...
  5. category is free text for organizing the catalog. We left version out of the entry on purpose: it belongs in the plugin's own plugin.json, which wins when both declare one.

We checked it with claude plugin validate ./team-plugins --strict on the 2.1.238 we have installed, which already ships that command. The first run failed: under --strict, a plugin.json without version counts as an error. Once the version moved there, the output ended in ✔ Validation passed. With no owner, the validator answered owner: Invalid input: expected object, received undefined, and with ./../plugins/reviewer as the source, Path contains "..". The line for your team's README is then claude plugin install reviewer --marketplace your-org/your-repo.

Security reviews: what 2.1.292 ships and what is still on you

Installing in one command also means the marketplace gets added without a pause in between. That makes it worth knowing what Claude Code checks and what it does not. This is what 2.1.292 brings on that front:

  • The same organization policies. A new marketplace added through --marketplace goes through the strictKnownMarketplaces allowlist and the blockedMarketplaces blocklist, exactly as with marketplace add. Per the page on how to manage plugins for your organization, both lists apply before anything downloads and again at every session start.
  • Policy loads first. On a first run, marketplace add and install no longer run before the managed settings arrive.
  • Look-alike names. Clearer steps when a marketplace declared in settings looks like an Anthropic one. Official names such as claude-plugins-official are only accepted when the source lives under github.com/anthropics/.
  • Security fixes. Among them: PreToolUse hook approvals and auto mode were bypassing the permission prompt for file reads from network (UNC) paths, and sandboxed commands could read the staged copies of /ultrareview uploads.

What it does not do is review the plugin's code. The plugin security and trust page says so in its first line: a plugin can execute arbitrary code on your machine with your user privileges. Its hooks, monitors and MCP servers run outside the sandbox, and your permission rules cover the tool calls Claude makes, not the code a plugin runs on its own. The docs suggest reviewing every plugin before you install it:

  1. Run claude plugin marketplace list to see where each marketplace came from.
  2. In a session, open /plugin, select the plugin and read its Will install section: commands, agents, skills, hooks, and MCP and LSP servers.
  3. Read the code: hooks/hooks.json (the command each hook runs), .mcp.json (each server's command or URL) and every file in bin/.
  4. Clone the repository and run claude --plugin-dir <directory> plugin details <name> to print its component inventory without starting a session.

With the one-step command, that review is yours to do before you paste it, because the shell will not ask. If you also want to scan the skills a plugin ships, see how to scan agent skills with SkillSpector. And if you were looking for a security review of your own code, that exists separately and did not change in this release: /security-review analyzes your branch's diff, and the security-guidance plugin reviews changes while Claude writes them.

Subagents with an effort level

Effort controls how much the model reasons on each step: lower is faster and cheaper, higher goes deeper. The custom subagents page gives two ways to set it for a subagent, and 2.1.292 adds the second one:

  • In its definition, with the effort frontmatter field. It applies every time the subagent runs.
  • On the call, when you ask Claude to run a (non-fork) subagent at a specific level. Claude passes the effort parameter, which overrides the field and stays in effect if the subagent is resumed.

The CLAUDE_CODE_EFFORT_LEVEL environment variable outranks both. Here is a reviewer subagent with its own level:

---
name: reviewer
description: Reviews the diff before a pull request. Use it after changes to business logic.
tools: Read, Grep, Glob
model: sonnet
effort: high
---

Read the changed files and report logic errors, uncovered edge cases and missing tests. Do not edit files.
  1. name and description are the only required fields. Claude reads the description to decide when to delegate.
  2. tools: Read, Grep, Glob leaves the reviewer with no editing tools.
  3. model: sonnet picks the model; without it, Claude Code follows its subagent model order.
  4. effort: high sets the level while the subagent runs, whatever level the session is on. A call that passes effort changes it for that invocation only.
  5. The body, after the second ---, is the subagent's system prompt.

Save it as .claude/agents/reviewer.md, or inside a plugin under agents/: plugin agents accept effort too. For the full walkthrough of a read-only reviewer, see how to build a code reviewer subagent in Claude Code.

The levels depend on the model, per the model configuration docs. Fable models, Opus 5.5, Sonnet 5.5, Haiku 5.5 and the others in that table accept low, medium, high, xhigh and max; Opus 4.6 and Sonnet 4.6 have no xhigh. If you ask for a level the model lacks, it falls back to the highest one it supports at or below it: xhigh runs as high on Opus 4.6. To see what actually happened, /tasks shows each subagent's model and the level set for it.

which effort level to give each subagentlevelwhich taskdefault onlowsmall changes you review: a rename—mediumday-to-day work with a clear scopeOpus/Sonnet/Haiku 5.5highfixing a bug in an existing codebaseother modelsxhighdeeper reasoning, more tokensOpus 4.7maxhard problems: finding vulnerabilities—who wins: the first one setCLAUDE_CODE_EFFORT_LEVEL>per-call effort (2.1.292)>effort in the filenone set: on the session's model, the session level appliesthe levels on offer depend on the modelsource: model-config and sub-agents docs, read on Nov 11, 2026

Scroll the figure sideways to see all of it

With no effort in the file or on the call, a subagent on the session's model runs at the session level. On another model, that level does not apply, and the docs say to set it in the file. The docs' level table is a sensible starting point: low for small changes you review yourself, high for fixing a bug in an existing codebase, and max only for hard problems, since the docs warn about diminishing returns and suggest testing it before adopting it broadly. A reviewer reading a large diff is a candidate for high; one that renames variables, for low.

How to update and check your version

All of the above depends on the version you run. These commands come from the Claude Code setup guide and the commands reference; we ran the first two on our machine, which has 2.1.238:

claude --version
claude plugin install --help | grep -- --marketplace
claude update
  1. claude --version prints your version. Here: 2.1.238 (Claude Code). You need 2.1.292 or later for --marketplace in the shell, and 2.1.275 for the session form.
  2. claude plugin install --help lists the options your version supports, and grep looks for the new one. On 2.1.238 it printed nothing: the help only lists --config, --scope and --yes. In PowerShell, swap the grep for Select-String -SimpleMatch "--marketplace", which also found nothing.
  3. claude update installs the newest version without waiting for the background update. Per the guide, it reports Successfully updated from <old version> to version <new version> or Claude Code is up to date (<version>). We did not run it: the rule for this test was to leave the install alone.

Native installs update themselves in the background. Homebrew and WinGet installs do not by default: update them with brew upgrade claude-code (or claude-code@latest, depending on the cask) and winget upgrade Anthropic.ClaudeCode. The autoUpdatesChannel setting picks between "latest", which gets each release as it ships, and "stable", which usually runs about a week behind. As of November 11, the npm latest tag points to 2.1.322 and stable to 2.1.311, so both already include what 2.1.292 added. If something looks off, claude doctor shows the result of the last update attempt.

the plugin commands you will use, and since which versioncommandwhat it doessinceclaude --versionprints your version—claude updateupdates now, no waiting—claude plugin marketplace add <src>registers a catalog—claude plugin install <p>@<mkt>installs from an added catalog—claude plugin install <p> --marketplace <src>adds the catalog, then installs2.1.292/plugin install <p> --marketplace <src>same in a session, asks first2.1.275claude plugin validate <path>validates marketplace, plugins—claude plugin marketplace remove <name>removes it and its plugins—in the shell, --marketplace needs 2.1.292 or later— = available earlier · claude plugin install --help lists your options

Scroll the figure sideways to see all of it

Updated November 11, 2026: more changed in this area after 2.1.292. Version 2.1.295 (October 8) warns when plugin install or marketplace add write to a settings file that does not load, and makes claude plugin validate suggest a README install line when one is missing. Version 2.1.296 (October 9) fixed installs from a GitHub owner/repo on Windows machines with no GitHub SSH key. Version 2.1.316 (October 31) added the CLAUDE_CODE_SUBAGENT_EFFORT variable for the level subagents run at when nothing else sets one; it is not on the environment variables page yet. And 2.1.317 (November 3) made a subagent on a different model use that model's saved level or its default instead of the session's.

Common mistakes

  • Writing plugin@marketplace together with --marketplace. With the flag, the plugin name goes alone: deploy-helper --marketplace acme/tools.
  • Using the repository name after the @. The marketplace name comes from its marketplace.json; acme/tools may register as acme-tools or as anything else.
  • Trying the flag on an old version. In the shell it needs 2.1.292. Check claude --version before blaming the marketplace.
  • Pasting a README command without reading the plugin. The shell adds the marketplace without asking. Review the plugin first, with the steps from the security page.
  • Expecting --scope project to share the marketplace. The marketplace goes into your user settings; each teammate runs the install in their own clone.
  • Typing the source without https://. gitlab.example.com/group/repo.git is read as a GitHub owner/repo and rejected.
  • Putting ../ in a plugin source. Paths resolve from the marketplace root, not from .claude-plugin/, and validate rejects them.
  • Giving every subagent effort: max. The docs reserve it for hard problems and suggest testing it first; and CLAUDE_CODE_EFFORT_LEVEL overrides it anyway.

Frequently asked questions

Do I need to add the marketplace before using --marketplace?

No. That is the point of the flag: if you do not have the marketplace, it adds it and then installs. If you already added it from the same source, it is reused. You only need 2.1.292 or later in the shell, or 2.1.275 or later inside a session.

Does 2.1.292 review the security of the plugins I install?

It does not review their code. It applies your organization's allowlist and blocklist, refuses names that imitate official marketplaces, and ships security fixes. Reading the hooks, the MCP servers and the files in bin/ before installing is still your job.

How do I ask Claude to run a subagent at a different effort level?

Say it in the prompt, for example: use the reviewer subagent at low effort. Since 2.1.292, Claude can pass that level as a parameter on the call, and it overrides the effort in the file. If CLAUDE_CODE_EFFORT_LEVEL is set, that wins.

How do I share a plugin with my whole team?

Publish the marketplace in a repository they can reach and put the --marketplace line in the README. To enable the plugin for a project, install it with --scope project and commit .claude/settings.json; each person runs the install once in their own clone.

Checklist

  • claude --version shows 2.1.292 or later.
  • claude plugin install --help lists --marketplace.
  • You read the plugin's hooks, .mcp.json and bin/ before installing it.
  • You know which source each marketplace was added from (claude plugin marketplace list).
  • Your marketplace.json passes claude plugin validate --strict.
  • The version lives in plugin.json, not repeated in the entry.
  • Each subagent has an effort chosen for its task, not max by default.
  • If your team uses policies, the marketplace source is in strictKnownMarketplaces.

If you have not built your first subagent yet, start with the read-only code reviewer, step by step. And if you are looking for something to install with the new command, go through the official Anthropic skills for Claude Code.

To keep plugins from eating your quota, read why your Claude Code usage limit runs out, and for project context, how AGENTS.md works in Claude Code.

Saw this on Instagram? Comment PLUGINS on the Reel and I'll send you the guide.

Written by

Luis Morales

Luis Morales

Frontend and Shopify developer, EcomLabs

@luismoralesdev on Instagram

Published November 11, 2026 · updated November 18, 2026

← Blog