What it is
Animal Protection Skills is a public GitHub repository of AI agent skills for people who work in animal protection. Vegan Hacktivists maintains it. Skills come from our team and from other organizations in the movement, and a member of our team reads every line of every skill before it goes in.
A skill is a folder holding one instruction file. Your AI agent reads that file when it decides the skill fits what you asked, then follows it. Skills teach an agent to do one job the way an experienced person would: critique a chart, strip the tells out of an AI-written draft, or catch you up on a project conversation you have not opened in months. If the phrase “AI agent” is new to you, What is an AI agent? covers the basics.
The repository uses the open Agent Skills format, so the same folder works in Claude (Code, Cowork, and claude.ai), OpenAI Codex, Gemini CLI, GitHub Copilot, and Microsoft 365 Copilot. You pick the agent; the skill stays the same. Set up Claude for your organization covers how skills work inside Claude itself, including how an admin reviews one before sharing it with the team.
What it is good at for advocacy orgs
You do not need to read the skills yourself. Paste the repository link into your agent and talk to it. Requests that work well:
- Finding what fits. “Read the skills index in this repository and tell me which skills would help someone who does fundraising for a farmed animal charity.”
- Checking before installing. “Read the graph-advisor skill and tell me whether it would help me. What does it need set up first?”
- Installing. “Install the de-ai-writing skill from this repository.”
- Adapting. “Install catch-up, but change it so the summary is written for my manager rather than for me.”
That last request is the point of a shared skills library. A skill is plain text, so your agent can read it, discuss it with you, change it to suit your organization, and install the changed copy. You are never stuck with our version.
Three skills are in the repository at launch:
- graph-advisor gives a blunt, specific critique of a chart against data visualization good practice, without redrawing it.
- de-ai-writing edits a draft to remove the patterns readers recognize as AI-written, changing as little as possible. If you would rather build your own version, Build a humanizer skill walks through it.
- catch-up summarizes a long conversation into what you asked for, where it got to, and what you need to decide next, in twelve sentences or fewer.
More arrive as the team and the movement submit them.
Where it fits
Most skills in the repository work inside whichever general assistant your organization already uses, such as Claude or ChatGPT through Codex. The repository does not replace those tools. It gives them a shared set of tested instructions so that two people at two organizations get the same quality of result from the same request.
It also gives your own work somewhere to go. If someone on your team has written a skill that saves them an hour a week, submitting it takes one pull request or one email, and they are credited by name in the skill and in the index.
Watch-outs
- A skill is a set of instructions your agent will follow. A badly written or malicious skill could tell an agent to do something you did not ask for. This is why every skill in the repository is read in full by a person before it is merged, and why we check for instructions fetched from elsewhere at run time, hidden text, unread scripts, and anything that asks for credentials or sends data where you did not intend. We still recommend you read a skill, or have your agent summarize it, before installing.
- Skills go stale. Command line tools change and connectors get renamed. Each skill carries a version number and a “last verified” date, and the index shows both. Treat an old date as a prompt to test before relying on it.
- Install location differs by agent. The instruction file is the same everywhere, but Claude Code, Codex, Gemini CLI, and Copilot each keep skills in a different folder, and Cowork and Microsoft 365 Copilot want a zip upload. The README lists the route for each. Your agent can usually handle this if you ask.
- The plain ChatGPT web app cannot install a skill folder the way this repository expects. ChatGPT Work can use skills too, through its own skill-creation flow or by pasting a skill’s file into a Project, and OpenAI Codex, on the desktop, command line, or in an editor, installs the folder directly. On a free or Plus ChatGPT account, a skill still works as pasted instructions, just without the one-command install.
- Check the requirements section. Some skills need a connector or a command line tool. Each skill says what it needs and how to get it on each platform. The three launch skills need nothing beyond the agent.
Getting started
- Open the skills index. It lists every skill with a one-line description, the author, what it needs, and when it was last checked.
- Pick one skill that matches something you already do each week. de-ai-writing is a safe first choice for anyone who drafts with AI.
- Give your agent the repository link and ask it to read that skill and explain what it would change about how it works with you.
- Ask the agent to install it, then run the skill on a real piece of work you were going to do anyway. Compare the result with what you would have done yourself.
- When you have a skill of your own worth sharing, follow the contribution guide in the repository, or email [email protected] with “Skill submission” in the subject line.
Pricing snapshot
Free. The whole repository is MIT licensed, which means any organization can copy, change, and redistribute the skills without asking. The agents that run the skills have their own pricing. What does AI actually cost? has the current figures.
