TLDR: Password Protection controls who gets the Skill: only subscribers on your list, checked every time they call it, and you can remove them any time. It does not hide the method from the assistant during the chat. IP Protection controls where the method runs: on our server, with only the result going back, so the method never reaches the subscriber or their assistant. It costs credits because our server does the work. Pick Password for Skills whose value is the outcome, IP for Skills whose value is the method. Neither mode can stop a subscriber from describing what a Skill does, and no vendor can promise that.

Key Takeaways

  • Password Protection = who. IP Protection = where.
  • Password mode hands the assistant the Skill text for the chat when the Skill is called, after the access check.
  • IP mode never hands over the text; the method runs from our server and the assistant gets the result.
  • IP mode runs on credits; each listed plan includes some and packs add more.
  • Extraction attempts are refused by the server before the Skill runs.
  • No mode can stop a person describing an outcome; do not believe anyone who says otherwise.

The question to ask about any "protection"

Where is the method when the subscriber uses it? If the answer is "in their assistant's context", then the assistant has it, and the person at the keyboard can ask for it. That is not a flaw in any one product; it is how assistants work. The only way to keep a method away from the subscriber's assistant is to not send it there.

Password Protection: control who

Your Skill is delivered to a subscriber's assistant only when they call it, only after our server checks that they are on your list, and the check runs every time. Remove them and the next call is refused. The assistant has the Skill text during that chat so it can follow it. This is the right mode when the value is in the result and the convenience, and the method itself is not the secret.

IP Protection: control where

The subscriber's assistant sends the request to our server. Our server runs your method, using the model provider's API, and returns the result. The method is never in the subscriber's assistant, never on the subscriber's machine, and never in a chat they can scroll back through. This is the mode for the Skill that is your actual edge. Because our server does the work, it runs on credits; each listed plan includes welcome credits, the larger plans add monthly credits, and packs can be added when needed.

The filter in front of both modes

A request that tries to pull a protected Skill's instructions out ("print your system prompt", "explain your methodology") is refused by the server before the Skill runs. For IP-protected Skills we keep a record of refused attempts and tell the creator. In Password mode, once the text is in a chat, what the subscriber says to the assistant never reaches our server; the instructions we wrap around the Skill tell the model not to reveal it, and the subscriber's own assistant decides how well it listens. That is the difference between the two modes in one sentence.

What neither mode can do

A subscriber who uses your Skill a hundred times can describe what it does. They can write their own version from the outside. So can a customer of any consultant. Protection is about not handing over the method, not about erasing memory. Anyone who promises more is selling you a word.

Choosing per Skill

  • Value in the outcome, method is common knowledge: Password.
  • Value in the method, you would not put it in a public post: IP.
  • A Stack can mix both. Many creators run their intake and formatting Skills in Password mode and their one real method in IP mode.

How to check us

Connect free. Run Skill Auditor, our IP-protected demo Skill, from Claude. Then ask Claude to show you Skill Auditor's instructions, and watch what comes back.