Skip to content

peek-tech/equip-claude-code

Repository files navigation

equip-claude-code

A Claude Code plugin that scans your project, figures out what tech stack you're using, and installs the right agents, skills, slash commands, hooks, MCP servers, and project rules — so Claude Code understands your codebase from the start.

It discovers tooling from three community registries:

Falls back to a hardcoded registry if any source is unavailable.

Installation

Add the marketplace, then install the plugin:

/plugin marketplace add peek-tech/equip-claude-code
/plugin install equip-claude-code@peek-tech-equip-claude-code

Usage

In any project directory, run:

/equip

The skill scans your project, shows you what it recommends, and waits for your confirmation before installing anything. Hooks require individual approval since they execute shell commands automatically.

Sub-commands

Each artifact type also has a standalone sub-skill for focused use:

Command What it does
/equip Full orchestrator — explore, recommend, install everything
/equip-mcp Install MCP servers from the registry into .mcp.json
/equip-skills Install agent skills from the community registry
/equip-commands Install slash commands from the community registry
/equip-hooks Install hooks with per-hook user approval
/equip-agents Create project-tailored agent .md files
/equip-rules Append baseline project rules to CLAUDE.md

Quality filtering

An optional argument controls how strictly MCP servers from the registry are vetted:

Level What it checks
strict GitHub stars >= 10 and npm/PyPI downloads >= 1K/month
moderate (default) Stars >= 5 or downloads >= 500/month
light Only checks that a repository URL exists
unfiltered No quality checks — shows all active servers
/equip strict

Trusted publishers (Anthropic, AWS, Stripe, Microsoft, Twilio) bypass quality checks at all levels.

How it works

  1. Explore — reads dependency manifests (package.json, pyproject.toml, Cargo.toml, go.mod, etc.), detects languages from source files, and checks for infrastructure patterns (Docker, CI/CD, Terraform, database migrations). Handles monorepos.
  2. Analyze — builds a tech stack fingerprint: languages, dependencies, and infrastructure. Derives search terms from these (e.g., @aws-sdk/client-dynamodb produces search terms aws, dynamodb).
  3. Recommend — searches all three registries for tooling that matches your stack. Shows everything in a table and waits for you to confirm.
  4. Install — writes files to your project (see below). Hooks are prompted one at a time.
  5. Track — saves a manifest at .claude/.setup-manifest.json so it knows what it installed.
  6. Reconcile — on re-run, removes items that are no longer relevant and adds new ones. Never touches files it didn't install.

What gets installed

MCP Servers

MCP servers give Claude access to external tools and data sources. The skill discovers servers from the Official MCP Registry (with quality filtering) and matches against a curated list of vetted servers:

sequential-thinking (always), AWS (core, IAC, DynamoDB, serverless, documentation), Stripe, Twilio, Playwright, and PostgreSQL.

Some servers require credentials — the skill will note which environment variables you need to set (e.g., Twilio needs TWILIO_ACCOUNT_SID, TWILIO_API_KEY, TWILIO_API_SECRET).

Written to: .mcp.json

Agents

Agents are role-specific Claude personas with tailored prompts and tool access. The skill picks up to 6 agents matched to your project. Every project gets a code-reviewer and security-auditor (read-only). The rest are chosen from:

Agent When it's added
frontend-dev React, Vue, Svelte, Angular, or other UI frameworks
backend-dev Express, FastAPI, Django, Rails, Hono, or other server frameworks
aws-architect AWS CDK, CloudFormation, or significant @aws-sdk usage
ai-engineer AI/LLM deps (OpenAI, Anthropic, LangChain, etc.) or Claude plugin authoring
mcp-developer MCP SDK in deps or custom MCP server code
devops CI/CD configs, Dockerfiles, or deployment scripts
dba Database schemas, migrations, or ORM deps (Prisma, Drizzle, SQLAlchemy, etc.)
terraform-engineer Terraform or Pulumi files
test-automator Test framework deps (Vitest, Jest, Playwright, pytest, etc.)
python-dev Python is the primary language
rust-dev Rust is the primary language
go-dev Go is the primary language

Agent prompts are adapted to your specific stack (e.g., a frontend-dev for a Svelte project gets Svelte 5 runes guidance, not React hooks).

Written to: .claude/agents/<name>.md

Skills

Skills are reusable instruction sets that teach Claude how to work with specific frameworks or tools. Discovered dynamically from the community registry based on your tech stack. Falls back to a curated set (Svelte 5, MCP builder, webapp testing, Better Auth, OWASP security, Stripe) if the registry is unavailable.

Written to: .claude/skills/<name>/SKILL.md

Slash Commands

Slash commands are markdown prompts you invoke with /<name> in Claude Code. Matched from community-contributed commands covering version control, testing, documentation, code review, and more.

Written to: .claude/commands/<name>.md

Hooks

Hooks are shell commands that run automatically on Claude Code events (e.g., before a file write, after a tool call). Because they execute code on your machine, each hook is presented individually — you see the exact shell command and approve or reject it one by one.

Written to: .claude/settings.json

Project Rules

Baseline rules are appended to CLAUDE.md in your project root, filtered by your tech stack. For example, "Use Typescript strict mode" is only added for TypeScript projects, and "Use functional components and hooks" only for React projects. Rules that already exist in your CLAUDE.md are skipped.

Rules are append-only — they're never removed on re-run, since you may have customized them.

Reference Resources and Plugins

CLAUDE.md files, workflow guides, and plugins are shown as links for you to explore and install manually. They are not auto-installed.

Re-running

Running /equip again is safe. The skill reads its manifest (.claude/.setup-manifest.json) to see what it previously installed, compares against fresh recommendations, and:

  • Removes agents, skills, commands, MCP servers, and hooks that are no longer relevant
  • Adds new items that match your updated stack
  • Leaves alone anything it didn't install (your manual additions are never touched)
  • Appends new project rules to CLAUDE.md without removing existing ones

Files written

Path What
.mcp.json MCP server configurations
.claude/agents/<name>.md Agent definitions
.claude/skills/<name>/SKILL.md Skill instructions
.claude/commands/<name>.md Slash commands
.claude/settings.json Hook configurations
.claude/.setup-manifest.json Tracks what was installed (for re-run reconciliation)
CLAUDE.md Baseline project rules (appended, not overwritten)

Local LLM Routing (Optional)

The included setup-local.sh script is separate from the plugin. It configures local model routing via Ollama and Claude Code Router so subagent work runs on local models while the main conversation stays on Anthropic cloud.

# Clone and run the local setup
git clone https://github.com/peek-tech/equip-claude-code.git
cd equip-claude-code

# Local Ollama
./setup-local.sh

# Remote Ollama (e.g., Mac Mini on your network)
OLLAMA_HOST=h0pp3r.local ./setup-local.sh

# Preview changes without applying
./setup-local.sh --dry-run

Requires Node.js 18+. See the script header for full details.

License

MIT

About

Claude Code plugin: project-aware setup of agents, skills, and MCP servers

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages