← Notes ·

My Claude Code Stack

The plugins, skills, and MCP servers I run day to day — and what each one does.

Claude Code is my primary coding environment. Over the last year it has grown from a chat wrapper into a full operating layer — planning, building, testing, shipping, and running my startup all happen through it. This is a clean tour of what I have wired into it and what each piece does.


Plugins

everything-claude-code

A massive curated library of coding standards, patterns, reviewers, build-error resolvers, and domain skills. It gives Claude opinionated references for any language, framework, or stack, plus workflow pieces like security reviews, test-driven development, and agent engineering. Think of it as a shared brain that raises the floor on everything Claude writes.

superpowers

Process discipline. Instead of letting Claude jump straight to code, this plugin enforces brainstorm → plan → test → implement → verify. Skills include brainstorming, writing-plans, executing-plans, test-driven-development, systematic-debugging, verification-before-completion, and using-git-worktrees. Rigid by design — you follow it, you do not negotiate with it.

claude-mem

Persistent memory across sessions. Claude normally forgets everything between conversations; claude-mem maintains a database of decisions, discoveries, and lessons so future sessions pick up where the last one left off. /mem-search finds prior context, /timeline-report narrates project journeys over weeks.

ui-ux-pro-max

Design intelligence for web and mobile. Seven skills covering design systems, UI styling, brand, banners, and slides. Opinionated enough that output does not look like another generic Tailwind template.

skill-creator

Meta-plugin for building and editing your own skills. When you notice yourself pasting the same prompt across projects, this is the tool that turns it into a reusable skill.

caveman

Output compression mode. Drops articles, filler, pleasantries, and hedging from Claude's text — cuts roughly 75% of output tokens without losing technical substance. Still writes normal English for code, commits, and security warnings.

gstack

Browsing, QA, and shipping toolkit. Provides /browse (headless Chromium for dogfooding), /qa (systematic app testing), /review/ship/land-and-deploy (full pipeline), /canary (post-deploy watch), /benchmark (perf regression), and /investigate (four-phase root-cause debugging). Installed globally so every project gets it.

GSD — Get Shit Done

A project-structure system. Eighty-plus commands for running real work with a paper trail. The core loop is /gsd-new-project/gsd-discuss-phase/gsd-plan-phase/gsd-execute-phase/gsd-verify-work/gsd-ship. Each phase gets its own directory, plan, atomic commits, and verification checklist.


MCP servers

MCP servers are bridges out of Claude into real systems. Each one gives Claude a set of tools that hit an external API.

Productivity

Docs and research

Dev tools

Persistent memory


Skills — auto-invoked (daily drivers)

These load by default and Claude matches them to context on its own.

Writing code

When editing Python, TypeScript, or Next.js, pattern skills like python-patterns, python-testing, pytorch-patterns, nextjs-turbopack, frontend-patterns, backend-patterns, api-design, postgres-patterns, database-migrations, docker-patterns, deployment-patterns, git-workflow, hexagonal-architecture, architecture-decision-records, and coding-standards quietly shape the output.

Testing

AI and agent engineering

Security

Research

Product

Ops (always on)

Workflow (gstack + GSD core)

Process + memory


Skills — on-demand (manual /invoke)

These are loaded but hidden from Claude's auto-match to keep the session context clean. I call them by name when the situation shows up.

Business

Ops

AI and agent specialty

Meta and config

Planning extras

Plan reviews

Design

Gstack extras

GSD extras

Security extras


Modes I run

Caveman

On by default in new sessions. Compresses Claude's output. Off with stop caveman or normal mode.

Auto-memory

File-based memory system at ~/.claude/projects/.../memory/. Per-project user, feedback, project, and reference notes. Separate from claude-mem.

CLAUDE.md router

Global instructions file at ~/.claude/CLAUDE.md. Tables map situations to the right skill to call. Keeps hidden / manual-only skills discoverable without loading them into Claude's auto-listing every session.


Why this matters

Software engineering is not getting slower. AI-native tools are becoming the primary interface, and the quality of your setup is becoming a real differentiator. The list above is not a flex — it is an attempt to build a personal operating system where the machine handles the work I should not be doing by hand.

The goal is not to install everything. The goal is to install exactly the things that earn their keep and route the rest out of sight until they are needed.