Claude Code setup

List of plugins, hooks and configuration settings I’m using in my Claude Code setup.

FYI I’m currently using the Max 5x plan, some settings could be adjusted differently for smaller or bigger plans

Plugins

All the plugins I’m using in my workflows. The general ones are always enabled and I activate the others per-project (e.g. the frontend ones only on frontend projects) or when I need them for a specific task, to avoid filling the context window with useless instructions.

General

• feature-dev: Structured workflow for feature development with specialized agents (review, architecture, security…). A good middle ground for tasks too complex for simple prompts & too easy for superpowers (more on that below).
• superpowers: Development workflow pushing Claude to ask you questions to really understand what you want him to build and writing design & implementation docs before proceeding with a subagent-driven-development process to work on it. Especially useful for complex problems with the brainstorm skill or when fixing issues with the systematic-debugging one.

Frontend

• frontend-design: Generate frontend interfaces that avoid generic AI aesthetics
• playwright: Gives Claude control of a browser so it can autonomously E2E test the changes it makes

Others

• context7: Up to date documentations for LLMs, useful when dealing with niche technologies or recent changes the LLM doesn’t know
• claude-dashboard: Configure a simple status line with all the useful information. Can be disabled after configuration to avoid clogging the context window.

Hooks

• RTK (Rust Token Killer): filters and compresses command outputs before they reach the LLM context, saving 60-90% of tokens on common operations. Very useful on the API plan to reduce your bill, and for plans like Max where you worry less about token consumption it’s still useful to have longer sessions before compacting.
• terminal-notifier, hook on Notification so Claude can nudge me when he needs my input while I’m working on something else

Permissions

I prefer to keep permissions scoped to the project level to avoid allowing too much by default, but I still have some basic allow and deny at the user level for easy of use & safety respectively.

• deny on Read(**/.env): .env files can sometimes be present in my projects, and while it never contains production secrets it’s better to avoid leaking it to Claude. In these cases I always have a .env.example file that it can read to learn what are the variables used.

Others

• Superwhisper: Local voice transcription that I use to talk to Claude Code when I need to better convey my thoughts (talking is also faster than typing)