Hands-On with Anthropic's Official claude-code-setup Plugin: Tailored Hooks and Skills Even for a Zero-Config Repo

When a Chrome extension project has almost no Claude Code configuration, can AI still offer genuinely useful automation suggestions? The author tested Anthropic's official claude-code-setup plugin on their own netflix-intro-skipper extension, and the results are worth noting.

What is the claude-code-setup plugin

claude-code-setup is an official, verified plugin published by Anthropic on the Claude plugin marketplace. Its purpose is to analyze a codebase and recommend project-tailored Claude Code automation configuration.

Its recommendations cover five categories: Hooks, Subagents, Skills, MCP Servers, and Plugins. A key design choice is that it is read-only—it never creates or modifies any files, so you can safely try it on an existing repository. By default it offers only 1-2 suggestions per category; when you explicitly request a specific category, it expands to 3-5.

Its method for determining the tech stack is no mystery: it reads files like package.json and the directory structure to infer the project type and toolchain.

The five carriers of Claude Code configuration: Hooks are shell commands triggered by lifecycle events (such as PreToolUse, PostToolUse) to enforce checks or block dangerous operations; Subagents are child agents with their own context windows and tool permissions, suited to isolating specific review tasks; Skills are instruction packages that encapsulate reusable workflows or domain knowledge, and can be set to user-triggered or autonomously invoked by Claude; MCP Servers connect Claude to external tools and data sources (such as browser automation or document retrieval) via the Model Context Protocol; and Plugins are the marketplace units that package and distribute the above capabilities. Only by understanding how these five divide responsibilities can you judge which form a plugin's suggestions should take.

Test subject: a nearly zero-config repository

netflix-intro-skipper is the author's own Chrome extension, supporting Manifest V3 and used to automatically skip Netflix intros. Its technical characteristics are as follows:

• Pure JavaScript, no build step
• Tests use Jest and fast-check (property-based testing), with 13 test files and about 6,000 lines of test code
• No CLAUDE.md; the .claude/ directory contains only a settings.local.json
• Spec documents live in .kiro/specs/
• src/content.js is a hefty 2,720 lines, the largest file in the project

Precisely because it has almost no Claude Code configuration, this project is ideal for testing the quality of the plugin's recommendations.

Installation and usage flow

Start Claude Code in the target repository and install the plugin from the marketplace:

/plugin install claude-code-setup@claude-plugins-official

After installation, you can confirm it on the Installed tab of /plugins. Then request an analysis with a single English prompt:

Analyze this project and recommend Claude Code automations (hooks, skills, MCP servers, subagents, plugins) that fit it.

The plugin returned a complete recommendation report, including a codebase profile, suggestions by category, and where to implement them.

Highlights from the recommendations

Hooks: designed around real pain points

The first suggestion is a PostToolUse hook: after each edit to a file under src/, automatically run npx jest --findRelatedTests to run only the related tests. The reasoning is straightforward—content.js has 2,720 lines, so even a small change could cause regressions, yet running the full test suite is too slow.

The second is a PreToolUse hook: block direct edits to version-management-related files (manifest.json, src/version.js, package-lock.json, coverage/). This is because the project syncs version numbers across three files via npm run version:*, and manual edits would cause them to fall out of sync.

Skills: codifying workflows and domain knowledge

The plugin suggests creating two Skills: release (user-triggered only, since releasing has side effects and requires dry-run, then version bump, then tagging) and netflix-dom-check (Claude-only background knowledge that summarizes the list of Netflix DOM selectors, the requirements in .kiro/specs, and the tests to check after updates). The latter targets the most common failure cause for this kind of extension—Netflix DOM changes.

Subagents: even historical PRs were factored into the analysis

The plugin recommends extension-security-reviewer (targeting the MV3 extension's extensive host_permissions, the XSS risks of DOM manipulation in content.js, and the lack of message-passing validation) and service-worker-reviewer. The basis for the latter is that PR #1 in the repository previously fixed a Service Worker issue after PC sleep, so it conducts targeted reviews of state persistence, alarms, and restart recovery.

MCP and plugins: filling the gap of real browser validation

The reason for recommending Playwright MCP is that Jest runs on jsdom and cannot validate the real popup UI and content script behavior; context7, meanwhile, provides the latest documentation for the Chrome Extension API, Jest, and fast-check. On the plugins side, commit-commands and frontend-design are mentioned, with the latter at lower priority.

How npx jest --findRelatedTests works: This flag makes Jest reverse-resolve the dependency graph based on Git-changed files and run only the test files that have import relationships with the changed files, rather than the entire test suite. For a project with a 2,720-line content.js and about 6,000 lines of test code, this can compress post-edit validation from minutes to seconds while still covering the directly affected cases. It relies on Jest's module dependency analysis, so coverage is limited for implicit dependencies established through dynamic require or global script injection—making it fast feedback rather than a full regression guarantee.

Key takeaways

The most valuable aspect of this report is that almost every suggestion can be traced back to concrete facts in the repository—file sizes, script logic, PR history. It does not offer a generic, one-size-fits-all checklist.

At the same time, note that the plugin only outputs suggestions and makes no changes. Implementation requires you to place the configuration manually or separately ask Claude Code to create it. The plugin's own first suggestion is to create a CLAUDE.md documenting how to run npm test, the version-sync rules, and the role of .kiro/specs; /init can generate a draft.

For projects unsure of where to start with configuration, this read-only, side-effect-free plugin is a low-cost starting point.

One-line Summary

Hands-on test of Anthropic's official read-only claude-code-setup plugin generating project-tailored automation suggestions for a zero-config Chrome extension

Paragraph Summary

The article tests Anthropic's official claude-code-setup plugin on the author's own Chrome extension netflix-intro-skipper. The plugin is read-only and modifies no files; it infers the tech stack by reading package.json and the directory structure, and offers suggestions across five categories: Hooks, Subagents, Skills, MCP Servers, and Plugins. Facing a nearly zero-config repository with no CLAUDE.md and only a settings.local.json, the plugin still produced highly tailored suggestions: a PostToolUse related-tests hook for the 2,720-line content.js, a PreToolUse hook to prevent manual edits to version files, a user-triggered-only release Skill, a Service Worker review subagent based on the PR #1 history, and Playwright MCP for real browser validation. The conclusion is that every suggestion traces back to concrete repository facts, but the plugin only outputs and does not implement; the first step it suggests is using /init to generate CLAUDE.md.