Free and open source

Stop re-explaining yourself to your AI

Your assistant isn't dumb. It just improvises a slightly different answer every time, because your standards live in your head. Skills are those standards, written down once in plain English, loaded automatically when your assistant recognizes the situation.

Copy the setup prompt and paste it into Claude. It asks what you want to be better at, shows you the few skills that fit, and waits for your yes before saving anything.

These are not a prompt roundup. They are the procedures running four AI agents that work every day on research, money, code, and communication. Each one loads the 13 to 18 of these that fit its job.

Set it up

One prompt. Nothing is saved until you approve it.

Uses Claude, which is a ChatGPT-style app with a free plan. ChatGPT can't do this yet. (Also Hermes, if you run your own agent.)

See exactly what you're copying

              

What's in the library

You don't need to pick from this list. The setup prompt asks what you want and suggests a few. This is just what's in there.

Every name links to the skill itself, so you can read exactly what it tells your AI to do before you install it.

Research and decisions

When you need to know something, or decide something.

deep-dive

“Go figure this out” comes back with a recommendation and what would change it, instead of a reading list.

Consults several different AI models on Hermes; on Claude it uses Claude alone.

mob-check

What real people are saying right now on Reddit, X, and YouTube, not the SEO version.

grok-search

Live web and X search when it needs what happened today.

Needs an xAI account

multi-review

Runs a draft, plan, or decision past several critical angles before you send it.

Reviews with several different AI models on Hermes; on Claude, Claude reviews its own work.

Getting work finished

For work that drags on, stalls, or piles up.

keep-going

It finishes the job instead of stopping to ask you which option you want.

project-steward

Runs a portfolio of long projects like a chief of staff.

Hermes only · Needs a Telegram bot

email-steward

Triages inboxes on a schedule and surfaces only what needs you.

Hermes only · Needs your email connected

address-pr-comments

For engineers: triages code review feedback and fixes what’s valid.

For developers, needs GitHub

pr-review-sweep

Nightly sweep of merged PRs for unhandled review comments.

Hermes only · For developers, needs GitHub

Making things

Turning messy input into something you can hand to a person.

google-docs

Turns messy notes into a real, formatted Google Doc.

Needs your Google account

google-sheets

Builds a real spreadsheet from raw data, formatted and ready to use.

Needs your Google account

google-slides

Turns an outline into a presentation deck.

Needs your Google account

diagram-rendering

Describe a diagram in words, get a real image back.

Some setup required

mini-app

Runs the little web apps your agent serves.

Hermes only · Server setup required

Reaching people

Your assistant talking to the outside world.

vapi-calls

It places an actual phone call: a booking, a confirmation, a reminder.

Needs a Vapi account

imessage-bluebubbles

Read and send iMessage from your assistant (Mac).

Mac only, some setup

report

File feedback from any session.

Hermes only

How your AI works

Its judgment, its memory, and its own upkeep.

trust-framework

Rules for when it acts on its own versus checks with you, and how it earns more rope.

recall

Picks up where you left off, across sessions.

Hermes only

memory-cleanup

Trims a bloated memory file without losing the facts that matter.

Hermes only

moa-solve

Throws a hard problem at several AI models and pulls out the best answer.

Hermes only

skill-librarian

Audits its own skills: duplicates, dead weight, and why one didn’t fire.

Written for Hermes' skill layout, so on Claude it advises rather than fully audits.

robustify-doctor

Tells you whether your agent is actually healthy.

Hermes only · For your own agent's setup

Before anything is installed

  • It asks what you want to be better at, then suggests a few skills rather than installing everything.
  • It shows you each one in plain English and where the files would live, then waits for your yes before saving or replacing anything.

Skills are readable text files, not programs, and your AI follows them the way it follows any instruction. Some can direct it to use accounts and tools you have connected, so read what it proposes before approving, the same as any AI output. Every skill is public on GitHub, and you can delete one at any time.

If Claude says it can't open the link

The Claude website and phone app can only open pages they found in a search, so they sometimes refuse a link handed to them directly. Claude Code and the desktop app don't have this limit. When it happens, install the file instead:

  1. Download a skill from the releases page. Each one is a small .skill file.
  2. In Claude, open Customize, then Skills, then Add.
  3. Upload the file. It stays available in every future chat.

Good ones to start with:

  • keep-going — finishes the work instead of stopping to ask which option you want.
  • mob-check — what real people say about something, not the SEO version.
  • trust-framework — decides when to act on its own and when to ask you first.

These three work fully in the Claude app and need no setup. Every skill in the library is on the releases page.

Ready to stop repeating yourself?

Tell it what you want to be better at. It suggests a few skills and shows them to you before anything is installed.