😎 Claude Vibes

A Claude Code plugin for people who describe WHAT they want and let Claude handle HOW.

It covers three things that Claude Code does not ship on its own:

• Build a brand from nothing. Sixteen guided commands take you from founder interview to a compiled brand guidelines document: purpose, values, positioning, archetype, voice, messaging, tagline, pitch, colors, and typography.
• Turn an idea into a plan, and stay on it. Discovery, scoping, and architecture, then the graph-engineering skill designs the whole project as a graph of jobs with real dependencies, separate checks, and approval points, covering code, brand work, content, account setup, legal, and launch. The approved graph is saved as a checklist that every command reads before it works, and anything not on it has to be added on purpose. That is what stops a project from drifting.
• A creator toolkit. Sponsor scripts, hooks, marketing copy, deep research, and image prompts for Midjourney and Nano Banana Pro.

It also wraps building, shipping, debugging, and refactoring in the same plain-language, confirm-before-acting style, and leans on Claude Code's built-in review, reasoning, and memory wherever those exist.

Installation

1. Install Claude Code

macOS, Linux, or WSL:

curl -fsSL https://claude.ai/install.sh | bash

Windows PowerShell:

irm https://claude.ai/install.ps1 | iex

This is Anthropic's native installer, and it keeps Claude Code updated automatically. Homebrew (brew install --cask claude-code) and WinGet also work, but they do not auto-update by default. Already have Claude Code? Skip this step.

2. Install Node.js 18+ (only for brand naming)

The Whois server, which checks whether a domain is free while you name a brand, starts through npx. If you will use that command, install Node from nodejs.org, or with brew install node on macOS. Everything else works without Node.

3. Install Claude Vibes

claude plugin marketplace add mike-coulbourn/claude-vibes

claude plugin install claude-vibes@claude-vibes

4. Start

Open a terminal in your project folder, run claude, and type / to see the new commands. If Claude Code was already running, restart it or run /reload-plugins.

To update later, run claude plugin update claude-vibes@claude-vibes, then restart Claude Code.

Local install (alternative)

git clone https://github.com/mike-coulbourn/claude-vibes.git

Add as a local marketplace:

/plugin marketplace add ./claude-vibes

Team installation

Add to your repository's .claude/settings.json for automatic team setup:

{
  "extraKnownMarketplaces": {
    "claude-vibes": {
      "source": {
        "source": "github",
        "repo": "mike-coulbourn/claude-vibes"
      }
    }
  },
  "enabledPlugins": {
    "claude-vibes@claude-vibes": true
  }
}

When team members trust the repository folder, the plugin installs automatically.

The workflow

BRAND → START → BUILD → SHIP → FIX → REFACTOR        + TOOLKIT (any time)
  ↓       ↓       ↓       ↓      ↓        ↓
Identity Plan   Code   Deploy  Debug   Evolve

Each phase has its own commands and agents.

Commands and agents

00-BRAND (brand identity)

Build a startup's brand identity step by step, from the founder interview to a finished guidelines document.

Commands:

Discovery phase

Strategy phase

Messaging phase

Visual phase

Compile phase

Agents:

• brand-archetype-selector - Jungian archetype selection
• brand-audience-researcher - Psychographic analysis and audience profiling
• brand-color-strategist - Strategic color palette development
• brand-competitive-auditor - Competitor visual identity analysis
• brand-elevator-pitch-writer - Elevator pitch variations
• brand-messaging-architect - Value proposition and brand pillars
• brand-naming-specialist - Strategic name development
• brand-positioning-strategist - Positioning and onliness statements
• brand-purpose-architect - Purpose, mission, vision (Golden Circle)
• brand-tagline-creator - Memorable tagline creation
• brand-typography-curator - Typography system and font selection
• brand-values-curator - Core values discovery and articulation
• brand-visual-director - Visual identity direction and briefs
• brand-voice-architect - Voice and verbal identity guidelines
• startup-brand-architect - Orchestrates complete brand identity creation

01-START (discovery and planning)

Plan before you build. Discover the problem, scope the MVP, design the foundation, and write the roadmap that the rest of the project follows. The roadmap lives in docs/01-START/roadmap.md in your project, so you can read it, edit it, and keep it in git.

Commands:

Agents:

• market-validator - Research market viability and competition
• feature-brainstormer - Generate and prioritize feature ideas
• tech-advisor - Recommend technology choices
• plan-reviewer - Review and improve implementation plans
• data-modeler - Design data structures and schemas

How planning and tracking work

1. Run the first three START commands. They produce plain-language documents on the problem, what is in and out of scope, and the technical foundation.

2. Run 04-plan-roadmap. It hands all of that to the graph-engineering skill, which first confirms an alignment contract with you (the objective, what done means, what to optimize for, what is out of scope, and which decisions stay yours), then designs the work as jobs. Each job has one owner and one checkable output. Reviews are separate jobs, so no job grades its own work, and an approval point sits in front of anything expensive or hard to undo. The skill only designs. Nothing runs until you approve the graph.

3. The approved graph becomes docs/01-START/roadmap.md, a file in your project that you can read and keep in git. One line per task:

   - [ ] 2.1 Build checkout | type: build | how: 01-plan-code, then 02-write-code | depends on: 1.4 | done when: a test card payment succeeds
   - [ ] 2.2 Review checkout | type: check | how: 03-review-code | depends on: 2.1 | done when: review passes with no blockers
   - [ ] 2.3 Open the Stripe account | type: setup | how: you, by hand | gate: your approval before starting | done when: live keys are saved
   

4. Work from the roadmap. 05-track-progress shows where things stand and moves the next task forward, whatever its type. For code, 01-plan-code picks the next build task and 02-write-code ticks it off once its done-when line is true.

5. Change course on purpose. Ask for something that is not on the roadmap and the command stops to ask where it belongs: a phase, the "Later" list, or "Not doing". A change to the objective itself sends you back to 04-plan-roadmap to approve a new graph.

Graphs are not only for roadmaps. graph-engineering works for any piece of work with several steps or sources, parallel paths, checks, risk, or approvals, and the skill itself tells you when a task is too simple to need one. Twelve commands offer it when the work is big enough: audience research, competitor audits, naming, and positioning in BRAND; discovery and architecture in START; planning and reviewing code in BUILD; diagnosing a bug with several possible causes; assessing a risky refactor; and the research and business-growth commands in the TOOLKIT. Once you approve a graph, graph-running carries it out at the pace the graph recommends, from walking you through one job at a time to running independent lanes as parallel agents. A separate agent plays skeptic so no job grades its own work, the work stops at every approval point, and results are saved to docs/graphs/. You can also start either skill directly with /claude-vibes:graph-engineering for anything else, such as a security review or a vendor decision.

TOOLKIT (specialized tools)

Commands and agents for work outside the development workflow: research, copy, scripts, and image prompts.

Commands:

Agents:

• brand-researcher - Deep brand research for content creators
• business-growth-advisor - Strategic business growth consultation
• deep-researcher - Research a topic in depth and report what experts actually do
• elite-copywriter - Write or rewrite marketing copy
• hook-generator - Write several opening hooks for a video
• nano-banana-pro-expert - Guidance for Nano Banana Pro image generation
• sponsor-script-writer - Write scripts for sponsored and affiliate videos

02-BUILD (implementation)

Build features methodically. Plan each feature, implement with best practices, and review before shipping.

Commands:

Agents:

• code-architect - Design feature architecture and patterns
• code-guru - Implement features with best practices
• code-reviewer - Thorough code review and quality checks
• tester - Write and run tests iteratively to prove code works

03-SHIP (deployment)

Check, commit, push, and open a pull request. The commit command confirms its message with you first, and nothing force-pushes without your approval.

Commands:

04-DEBUG (debugging)

Fix bugs systematically. Diagnose root causes, apply fixes, and verify they work.

Commands:

Agents:

• diagnostician - Deep investigation and root cause analysis
• fixer - Minimal, targeted bug fixes
• verifier - Verification and regression testing
• tester - Write and run tests to prove fixes work

05-REFACTOR (improving existing code)

Improve code without changing behavior. Assess opportunities, refactor safely, and validate preservation.

Commands:

Agents:

• assessor - Code archaeology and improvement analysis
• refactorer - Safe structural improvements
• validator - Behavior preservation verification
• tester - Write and run tests to prove behavior preserved

Skills

Skills are knowledge packs that Claude loads on its own when a conversation calls for them, in any conversation once the plugin is installed. The commands and agents above rely on them. You can also run one directly, which is how you would normally start the two thinking tools: /claude-vibes:interview-me and /claude-vibes:graph-engineering.

platform-optimization and midjourney-prompting describe fast-moving targets. Each carries a last-verified date and tells Claude to check current sources before relying on specific numbers.

Building your own skills, agents, hooks, or plugins? Version 1 shipped builder skills for that. Anthropic now maintains better ones alongside the product, so install those instead: /plugin install plugin-dev@claude-plugins-official and /plugin install skill-creator@claude-plugins-official. Claude Code adds that marketplace automatically. If it reports the marketplace as not found, run /plugin marketplace add anthropics/claude-plugins-official first.

MCP servers

MCP (Model Context Protocol) servers give Claude Code extra tools. The plugin installs two that need no setup.

Installed automatically

These servers start automatically when the plugin is enabled:

Notes:

• Add use context7 to a prompt to pull current library documentation. Context7 is a hosted service, so those lookups leave your machine.
• Whois runs locally through npx, pinned to an exact version.
• Project tracking needs no server. The roadmap is a file in your project that the commands keep up to date.
• Cross-session learning uses Claude Code's native agent memory, stored in .claude/agent-memory/ in your project, so there is no memory server. Commit that folder to share it with your team, or add it to .gitignore to keep it private.

Optional servers (setup required)

None of these are needed by the plugin. They are the ones that pair well with it, and each asks you to sign in the first time you use it (run /mcp inside Claude Code). All use the HTTP transport, since Claude Code has deprecated SSE.

Vendors change these endpoints, so if one fails, check the vendor's MCP page or the Claude Code MCP guide.

Keep the list short

Every MCP server adds to Claude Code's startup time, so add one only when you need it.

Structure

claude-vibes/
├── .claude-plugin/
│   └── marketplace.json         # Marketplace manifest
├── .github/                     # CI workflow, issue and PR templates
├── scripts/
│   └── validate_plugin.py       # Checks frontmatter, references, links, and counts
├── CHANGELOG.md
├── CONTRIBUTING.md
├── SECURITY.md
├── CLAUDE.md                    # The vibe coding framework this repo is built with
└── plugins/
    └── vibes/                   # The plugin
        ├── .claude-plugin/
        │   └── plugin.json      # Plugin manifest, including MCP servers
        ├── commands/
        │   ├── 00-BRAND/        # Brand identity
        │   ├── 01-START/        # Discovery and planning
        │   ├── 02-BUILD/        # Implementation
        │   ├── 03-SHIP/         # Commit, push, pull request
        │   ├── 04-DEBUG/        # Debugging
        │   ├── 05-REFACTOR/     # Improving existing code
        │   └── TOOLKIT/         # Research, copy, scripts, image prompts
        ├── agents/
        │   ├── BRANDING/
        │   ├── CODING/
        │   └── TOOLKIT/
        └── skills/              # 22 skills

Contributing, security, and license

See CONTRIBUTING.md before opening a pull request, and CHANGELOG.md for what changed in each release. Report vulnerabilities privately as described in SECURITY.md. If a skill teaches something that is out of date, open an issue with the "Outdated or wrong content" form. Released under the MIT License.

References

• Plugins and the plugins reference
• Skills and custom commands, which are now one system
• Subagents
• MCP in Claude Code
• Claude Code overview and best practices