Repos I actually use. With warnings.

anthropics/skills

Anthropic's library of Agent Skills: folders of instructions and scripts Claude loads when the task fits.

If you have typed the same instructions to Claude twice, it should be a skill. Start by reading how these are written.

anthropics/claude-code

Claude Code, the agentic coding tool that works in your terminal and editor.

Most of the custom agents I build start life here: a folder, a few files, a clear brief.

modelcontextprotocol/servers

Reference servers for the Model Context Protocol: the plumbing that connects AI to your apps.

Understanding connectors is the difference between a chatbot and an assistant that can actually do things.

NousResearch/hermes-agent

An open-source agent you can run yourself, with memory, skills and tool use.

My own agent fleet runs on Hermes profiles. Good for owners who want agents on hardware they control.

Watch out: Give it the narrowest access that works, and read what each skill can touch.

openclaw/openclaw

A personal AI agent that runs locally and acts through your apps and browser.

A fast way to see what a local agent can do on a real machine.

Watch out: It asks for broad system permissions. Use a separate account or machine for experiments.

n8n-io/n8n

Workflow automation you can self-host, with AI nodes and hundreds of integrations.

When a job is a clear sequence of steps, a workflow beats an agent. Cheaper, easier to debug.

D4Vinci/Scrapling

A Python scraping library with adaptive selectors and a stealthy fetcher for public pages.

I used it for public-page research where simple requests break.

Watch out: Respect each site's terms and rate limits. Public data only.

microsoft/playwright

Browser automation for testing and for reading pages that only render with JavaScript.

The fallback engine inside my read-only social media extractor.

ollama/ollama

Run open models locally with one command.

Cheap triage and private batch jobs stay on your own machine. Save the expensive model for the hard cases.

supabase/supabase

Postgres database, auth and storage with an open-source stack.

My default backend for small agent apps that need a real database and row-level security.

resend/react-email

Build clean, responsive emails as React components.

For agents that send reminders or digests, emails that look like a real business sent them.

huggingface/agents-course

Hugging Face's free course on AI agents: fundamentals, frameworks, tools and reasoning.

The best free place to learn how agents actually work before you trust one with your business.

anthropics/courses

Anthropic's educational courses, including prompt engineering and tool use.

Frameworks before tools. These teach the habits that survive every model release.

hammem/monarchmoney

An unofficial Python library for Monarch Money's personal finance data.

A good example of why I run a security review before connecting anything to money.

Watch out: It handles your financial login. Never hard-code credentials; review session storage first.