THE MASKED FOUNDERANONYMOUS BUILD LOG
The toolbox . repo shelf

The repos I actually
build with.

8 open-source projects that earned a place in my real work. What each one does, why I trust it, and the safe way to try it. Built for Claude Code, in plain English.

Check before you install. Every time.

A repo is a stranger's code on your machine. Before you add ANY of these, paste the repo into the three buckets prompt from the prompt library and make your AI sort its claims: proven, contradicted, or backed by nothing. Claims backed by nothing are worth zero until you test them yourself. I do this for every pick on this page, and you should do it to my picks too.

Start here

anthropics/claude-code

The AI worker this whole site is built around. It lives in your terminal, reads your project, and does real work end to end.

Why I trust it: Every prompt, agent, and skill in this library was made for it and tested in it.

Safe first step

  • Install it, open an empty folder, and ask it to build something small. One hour of doing teaches more than a week of reading about it.
Open on GitHub
Teach it to work

obra/superpowers

A pack of ready-made skills that teach your AI a careful way to work: plan first, test first, check before you claim done.

Why I trust it: My plan-first habit comes from here. These rules run on every build I do.

Safe first step

  • Read the brainstorming and debugging skills first. Even if you install nothing, the way of working is the lesson.
Open on GitHub
Teach it to work

anthropics/skills

Anthropic's own library of skills you can drop straight into Claude Code.

Why I trust it: I copy its structure every time I write a new skill. Official and well built.

Safe first step

  • Skim the skill files themselves. Notice how short and direct they are. That is the bar.
Open on GitHub
Give it tools

modelcontextprotocol/servers

The official set of connectors that let your AI reach real tools: files, databases, search, and more.

Why I trust it: When my AI needs a new ability, I look here before I build anything myself.

Safe first step

  • Browse the list and pick ONE connector you would actually use this week. Add one at a time, never ten at once.
Open on GitHub
Give it tools

upstash/context7

Feeds your AI the current docs for any library, so it stops guessing from old memory.

Why I trust it: It kills a whole class of wrong answers where the AI writes code for last year's version.

Safe first step

  • Add it, then ask your AI a question about a fast-moving tool you use. Compare the answer with and without it.
Open on GitHub
Let it check its work

microsoft/playwright-mcp

Lets your AI drive a real web browser: click, type, and see pages the way a person does.

Why I trust it: My AI checks its own web work with this before I ever look at it. That loop is the whole game.

Safe first step

  • Add it and tell your AI: build the page, then open it in the browser and prove it works before you show me.
Open on GitHub
Feed it media

m-bain/whisperX

Turns voice recordings into clean, timestamped text, all on your own computer.

Why I trust it: I use it when a recording is private. The audio never leaves my machine.

Safe first step

  • Record a one-minute voice note and run it through. Check the text against your own ears before trusting it on real work.
Open on GitHub
Feed it media

yt-dlp/yt-dlp

Downloads video and audio from almost anywhere, so your AI can study it.

Why I trust it: My research loop starts here: pull the video, read the words, keep the lessons, cite the source.

Safe first step

  • Use it on your own content first. Check each platform's terms before you pull anything, and respect creators: studying a source is fine, reposting it is not.
Open on GitHub
Go deeper

Tools are half the story. The system is the other half.

These repos give your AI hands. The skills library teaches it judgment: how to plan, check its own work, and hand off cleanly.

Open the skills library
Best Open-Source Repos for Claude Code . The Masked Founder