Who Owns AI Skill Files: Employer or Employee?

What AI skill files are, and why they are your most ownable AI work

“Skill files” are the human-written instructions that tell an AI how to behave: system prompts, prompt libraries, rules files such as a .cursorrules or a SKILL.md, agent workflows and loops, and the configuration that encodes how your tools should approach your work. They are the steering, not the output. A developer who has spent months refining how your AI handles your contracts, your code style, or your support queue has built something with real, transferable value.

Here is why they sit on stronger footing than AI-generated code. Under United States copyright law, copyright protects only what a human creates, which is what makes purely AI-generated output so hard to own. Skill files do not have that problem, because a person wrote them. The human-authorship barrier is cleared at the door. In my Minnesota business-law practice, this is the AI asset owners most often overlook and most clearly can protect. The question is not whether a human authored a skill file; it is which doctrine protects it, and whether that protection lands with the company or the person who typed it.

Copyright: you wrote it, so you can own it, within limits

A skill file is human-authored text, so it is eligible for copyright as a literary work, but copyright law removes the functional core of it. Under 17 U.S.C. § 102(b), copyright never protects “any idea, procedure, process, system, method of operation, concept, principle, or discovery.” A skill file is, in large part, a procedure: do this, then that, in this order. The merger doctrine (where an idea can be expressed in only one or a few ways) strips out more. What copyright actually protects is the original expression in the file, the particular wording, and the creative selection and arrangement of its parts, not the underlying method.

The practical consequence is that copyright in a skill file is real but often thin, and thinnest for short files. A two-line rule is mostly function and little expression. A large, distinctively organized prompt library has more protectable expression in how it is written and arranged. Copyright is worth securing, but for most skill files it is a backstop, the same role it plays for AI-generated code. The stronger protection comes from the next two doctrines.

Trade secret: the strongest protection for skill files

Trade secret law is the natural fit for a valuable skill file, because it protects codified know-how regardless of how functional it is, as long as you keep it secret. Under the Minnesota Uniform Trade Secrets Act, Minn. Stat. § 325C.01, subd. 5, information is a “trade secret” when it “derives independent economic value, actual or potential, from not being generally known to, and not being readily ascertainable by proper means by, other persons who can obtain economic value from its disclosure or use,” and “is the subject of efforts that are reasonable under the circumstances to maintain its secrecy.” A refined prompt library that gives your business an edge, and that competitors do not have, fits that definition. The functional-content exclusion that thins copyright does not apply here.

The catch is the second half of the definition, and Minnesota courts enforce it strictly. In Electro-Craft Corp. v. Controlled Motion, Inc., 332 N.W.2d 890 (Minn. 1983), the Minnesota Supreme Court reversed a trade-secret judgment for an employer because, even though its information had value, the employer “did not meet its burden of proving that it used reasonable efforts to maintain secrecy.” Value is not enough. If your skill files sit in an open repository that every employee and contractor can read, with no confidentiality terms and no access limits, you may have forfeited trade-secret protection before any dispute arises. Treat the files as confidential: access controls, need-to-know limits, and confidentiality terms with everyone who touches them. One more limit follows from the same body of law: a skill file that merely strings together publicly known prompts may be too generic to be a secret at all, because what is generally known in the field is not protectable.

Patent: almost never the tool

A patent is rarely the right tool for a prompt or a workflow. A patent protects an invention that is new and non-obvious, but the threshold question is subject-matter eligibility, and that is where prompts fail. Under 35 U.S.C. § 101, as the Supreme Court applied it in Alice Corp. v. CLS Bank International, 573 U.S. 208 (2014), an abstract idea does not become patent-eligible just because it is carried out on a generic computer. A prompt or an agent workflow is, in patent terms, close to a pure abstract idea, so it usually fails that test. Unless your team has built a genuine technical invention rather than a clever set of instructions, patent protection is not the play for skill files.

Employer or employee? Work-for-hire and the assignment gap

When an employee writes a skill file as part of the job, the employer owns it, but a contractor is a different story. Federal copyright law treats an employee’s in-scope work as a work made for hire: 17 U.S.C. § 101 defines it as a work an employee prepares “within the scope of his or her employment,” and 17 U.S.C. § 201(b) vests ownership in the employer, who “owns all of the rights comprised in the copyright” unless the parties agree otherwise in writing. So a skill file your engineer builds on company time, for company systems, is the company’s: the copyright by operation of that rule, and the trade-secret protection too, as long as you keep the file confidential.

Independent contractors are the gap, and it is a common one in AI work, where businesses bring in outside developers and consultants. Work made for hire does not cover an independent contractor’s work automatically. Absent an express written assignment, the contractor can own the skill files they built for you, and you may hold only a license to use them. The fix is straightforward and belongs in every engagement: an invention-assignment and confidentiality clause, signed before the work begins, that assigns to your business all rights in what the contractor creates. The same care that makes a standalone confidentiality agreement enforceable in Minnesota applies to the assignment language in a contractor agreement.

The developer who leaves: general skill versus your trade secret

The hardest question arrives when a developer who built your skill files leaves for a competitor. Minnesota law draws a line that cuts in both directions. An employee is free to use the general skill, knowledge, and experience gained on the job at a new employer, but may not take the former employer’s trade secrets. The Minnesota Supreme Court recognized that balance in Electro-Craft Corp. v. Controlled Motion, Inc., acknowledging the tension between an employer’s right to protect genuine secrets and an employee’s right to compete using the knowledge they have acquired. So whether a departing developer can reuse the skill files they wrote turns on three things: how the files were developed (on company time, for company systems, or on their own), whether you actually kept them secret under the reasonable-efforts standard above, and whether the developer signed an assignment and confidentiality agreement.

You cannot close this gap by simply barring the developer from competing. Since July 1, 2023, Minn. Stat. § 181.988 has made a covenant not to compete in an employment contract “void and unenforceable,” for agreements entered into on or after that date. A Minnesota noncompete signed with a current employee will not hold. But the same statute expressly preserves the tools that actually protect skill files: a covenant not to compete, it says, “does not include a nondisclosure agreement, or agreement designed to protect trade secrets or confidential information,” and does not include a nonsolicitation or customer-list agreement. The lesson for your skill files is direct. Your protection is not a noncompete; it is a confidentiality and assignment agreement backed by real secrecy. That is exactly the combination that lets you stop a former employee from taking confidential information, and it is why a properly drafted Minnesota nondisclosure agreement is worth more here than a noncompete ever was.

Which law protects your skill files

The four doctrines do different jobs, and the practical answer is to layer the ones that fit. The table below is the short version.

What to do now

Skill files are valuable enough, and portable enough, to warrant the same protection you would give any core intellectual property. The steps are concrete:

1. Assign and protect at hire. Put an invention-assignment and confidentiality clause in every employee and contractor agreement, signed before the work starts. Contractors especially, because they are not covered by work-for-hire. Include the federal Defend Trade Secrets Act immunity notice (18 U.S.C. § 1833(b)(3)) in any agreement that governs confidential information, because leaving it out forfeits exemplary damages and attorney fees in a later federal trade-secret suit against that person.
2. Name skill files as company work product. State in writing that prompts, rules, configuration, and skill files developed for the business belong to the business. Do not leave it to inference.
3. Keep them secret. Access controls, need-to-know repositories, and confidentiality terms, so the files meet the reasonable-efforts standard the Minnesota Uniform Trade Secrets Act requires. Avoid pasting them into AI tools whose terms let the vendor train on or retain your inputs.
4. Handle departures deliberately. When someone leaves, retrieve the files, disable access, and document it, so your secrecy measures hold up if the question ever reaches a court.
5. Register the strongest ones. For a distinctive, high-value skill file, register the human-authored expression with the U.S. Copyright Office as a backstop, disclaiming any AI-generated portions.

For the other half of this question, how much of your AI-generated code you can actually own, see the companion guide on who owns code that AI writes. If you would like to make sure your business owns and protects the skill files your team is building, email [email protected]. For the broader picture, see how I approach copyright and intellectual property work.