Reading Time: 7 minutes

My lawyering skills class turned in their first predictive memos and so we spent some time turning our attention to research. I have been mixing AI research in as we go and I finally turned to Skills. The more I prepared, the more I came to the conclusion that (a) Skills may be the most important thing that future lawyers understand about their AI usage and (b) I’m not sure why we need legal publishers involved in the workflow unless they’re going to provide value.

I still think Skills is a dumb term of art. One of the students asked me what I would call them if not Skills. To me, they’re templates. It’s not an edgy term but it better describes what they are: a customizable document that can be re-used in order to create standardization. Also, lawyers have used templates forever, whether paper work product or electronic forms.

Skills as Legal Technology

The development of the Skills “standard”—everything related to generative AI seems to require caveats and quotation marks—reinforces my view that AI is still half-baked. Yes, you can get it to generate things. The emergence of Skills shows a refinement towards productivity. The Skills “let you package domain expertise and organizational knowledge” including “procedural knowledge: workflows, best practices, and guidance” that you or your law firm has. It creates a container in which the AI can operate based on more than what the AI developer prioritizes. The Skill document captures knowledge or information or process that you want to make repeatable (without having to manually repeat it).

This ability to have something that is reproducible is key. I have earlier posted that I didn’t think that lawyers would tinker with Skills any more than they have done with macros. This is especially true if Skills are submerged beneath product cruft where their real functionality is hidden. However, given the low tech barrier to creating or modifying one, I’m starting to wonder whether, if lawyers can deploy them as easily as a browser extension or Outlook add-in, they might not offer more value as a legal technology.

We spent some time looking at the Thomson Reuters CoCounsel Library, which can be filtered by type to show just Skills. Similarly, we look at RELX Protégé and its library of Skills, which can be filtered by source, including those from Anthropic’s Claude and “Custom”. Unfortunately, Skills on the legal publisher sites are opaque. Even though a Skill is a text file written in Markdown (which is displayed by the file extension .md), the files themselves do not seem to be accessible from within the legal publisher platforms. As with so many things from the legal publishers, we are intended to take on faith that what they have provided is adequate.

Here’s Lexis’s explanation of their approach:

LexisNexis legal professionals have built a set of useful skills based on our extensive legal experience and thorough testing with customers. Users can also access skills that were developed by Anthropic, the makers of Claude. Users and administrators can create reusable skills based on their own preferred templates or processes. These skills can be shared across teams within the organization. Skills can be deployed for common and complex tasks, providing a guided starting point for activities such as contract review, document analysis, and drafting support. 

A New Way of Working: Building Legal Work Around the Matter, Lexis PR, August 24, 2026

At least for academics, the access stops at the first sentence. There are not currently any ways to add user-created Skills. At least on Lexis you can see that the functionality may be coming. On Westlaw, you get what you see.

But the power is in the latter part of that puff: the ability to create your own Skills and to share them. So it is a bit disappointing that the publishers themselves are opaque. There’s a link to Anthropic’s Claude plugin marketplace but that’s not very helpful.

What is far more helpful, and where I took the students next, was to the Claude for Legal plugin repository on Github. In the folders organized on this site, all of the plugins and the Skills on which they rely can be read. Students can understand what a Skill is and, if they run one, understand some of the choices it is making behind the scenes. (I have forked the repository so I can tinker with the docs and just in case things disappear in the future).

A screenshot of the CoCounsel-Legal plugin Skill for Westlaw US Deep Research

The folder named external_plugins/cocounsel-legal can be matched up with the Skills available on CoCounsel for “Deep Research”. I found it useful to see inside because you can see some of the use cases it will stop and throw an error on. This means you can invoke one of these use cases, get the AI chatbot feedback, and show students the connection.

If the request falls into one of the categories below, briefly explain that this skill isn’t the right fit and point the user to the suggested alternative.
Retrieving the full text of a specific document
Instead: Suggest the traditional search box on Westlaw.
Summarizing what a specific statute, regulation, or treatise says on its own (e.g., “what does the California Evidence Code say about hearsay?”)
Instead: Suggest the traditional search box on Westlaw.
Analytics requests (“How often has Justice Scalia ruled in favor of…?”) Instead: Suggest Litigation Analytics on Westlaw.
Calculations (“What is the last possible filing date if…?”)
Instead: This request is out of scope of Westlaw Deep Research
Outcome predictions
Instead: This request is out of scope of Westlaw Deep Research

Westlaw Deep Research Skill.md

I didn’t hunt through all of the other folders that were above this one for Thomson Reuters Skills, but Lexis Skills are also sprinkled across this repository. In class, with the Lexis Protégé list of Student Skills open, we could see the same list of plugins on the Claude repo. Now, in addition to running a Flash Cards Skill, we could see how the app was using the Skill to build the output.

A screenshot of the contents of a folder on the Claude for Legal Github repository

I wish this sort of clarity was available within the legal publisher platforms. I suppose that, if and when the custom Skills option becomes available, we might see a bit more transparency. In particular, I would love to start to see Skills that other law librarians create and share within their firms.

But my message to my class was really about them. We are moving away from the basics like how to prompt a generative AI. For one thing, the best practices—using personas, for example—change over time as models iterate and may be different depending on the purpose to which the AI app is being turned. Even if teaching this sort of literacy could be made sticky enough, like internet search, those skills will atrophy the further they get in time from the training.

If I was a law student in 2026 and thinking about how best to prepare for AI in law practice, it would be by understanding how Skills can be deployed. In this way, Skills-as-templates are the successors or siblings of macros and Microsoft Office custom templates, and other automation tools. A law student does not need to know the technology—you can make toast without knowing how to make a toaster—so much as how to use it.

Some of the law students were surprised that they could see the inside of the clock when we read through a couple of the Skills to see what made the machine tick. You could see from their comments and questions that they could see how re-usable guardrails or guidelines could help in a law practice. A law student who is conversant in how to write a Skill or, better, to edit an existing Skill to tailor it to their new employer’s workflows, has a pretty powerful … skill.

About the Legal Publishers

So what about the legal publishers? Let me be clear that I don’t suggest that the opacity on the Skills is nefarious. It does beg the question about the value publishers are providing in this particular scenario. In the same way that they take specific collections of public domain information and make it searchable, I was a bit disappointed that the Skill source was such a generic one. If the legal publishers are actually “listening to their customers”, how does that align with “we are using Skills developed by an AI company?”

What if we can use Skills outside of legal research databases, since that is where they are getting their Skills anyway? What is our ability to further remove ourselves from having to rely on closed commercial gardens of information? The big problem with generative AI and lawyers right now is the lack of guardrails and professional incompetence. The latter problem can be managed by lawyers learning to return to their professional obligations: read the law they cite, for one thing.

If Skills provide guardrails and repeatable process controls, is the need for legal publisher systems further weakened? I had a student come up to chat about their summer work and how they had found a legal answer on an AI platform that they had been unable to find on either Thomson Reuters or RELX platforms. This is probably more surprising to a student and perhaps a lawyer than it is to a librarian; we know that the commercial platforms are neither comprehensive nor consistent.

We may already be recommending that law students with a known citation go to a free resource—like the authoritative publisher, the court or legislature—to pull the document. Why incur a charge in a commercial database for something so rudimentary. Artificial intelligence is showing some value at the other end of the spectrum, where you can take an unknown legal domain plus some facts and retrieve a custom secondary source and have a starting point for your legal research. Who needs West Publishing and Matthew Bender texts at that point (and, given the legal publishers inability to provide any access data that their digital secondary content is ever used anyway, it may be reasonable to assume that it is unused in any event).

AI tools with Skills may be the way to manage the unknowns of the center of the search and retrieval spectrum, between the known cite and the unknown legal domain. The sweet spot where you have keywords and maybe a starter case but need to get further into the weeds. If Skills can provide guardrails for that, and public or public-adjacent sites (like CourtListener) can be tapped, it can further reduce the need for a commercial license.

AI is neither a replacement nor a tipping point towards one for some of the tools that remain primarily in commercial legal publisher platforms: citation tools are perhaps the most important. At the same time, there comes a point where the value proposition of a diminished set of resources undermines the price.

I was thinking this when I noticed that an academic Lexis user can run 60 Skills in a month and 25 Deep Research iterations. There’s clearly a cost sensitivity that we’re currently avoiding. In a conversation with a legal publisher, I understand that their normal “notional” pricing (fake numbers attributed to activities taken by researchers to enable bill back amounts for law firms) for artificial intelligence was zero: no cost. But we all know that AI comes with expenses. A Microsoft 365 family license is $39 more per year with the Copilot add ons. Just because it isn’t visible doesn’t mean someone isn’t quantifying it, in this case in the number of times an AI function like Skills can be invoked.

It is not hard to imagine that someone who can create their own Skills-infused environment who runs into a usage cap may move outside the restricted platform. If the only vendor lock-in is that distinction—people who have an AI tool that supports Skills and people that don’t—I’m not sure it’s a very strong impediment. Especially not if we can share and reuse Skills that even the legal publishers are themselves not creating from scratch.

In keeping with their lack of business analytics, the legal publishers continue to do themselves a disservice by not measuring academic usage. It would be meaningful to know how many law students are valuemaxxing and using all of their monthly Skills and on what Skill. In the same way that they omit usage data to help us understand whether their electronic secondary sources are used, or if anyone uses their citators, legal publishers miss opportunities to show that they have value that cannot be achieved without their product.

In the past, I have thought about how strong the demand for legal publisher platforms is. In particular, I remain a believer that we do not need to have multiple platforms if what we are teaching is literacy. Not only could law schools save substantial expenses by limiting themselves to a single platform, it becomes less and less clear what the distinguishing factors are that recommend keeping more than one.

If anything, the existence of something like Skills gives me even more belief that some of the free-range, open web research issues can be managed. Don’t get me wrong: I am not by any means persuaded that artificial intelligence is the future of legal research. However, it seems inevitable that we will continue to see its merger with search deepen. To the extent we are doing specialized search and retrieval and can use Skills to constrain how the opaque tools support our work, I welcome the additional certainty they might provide a researcher.

Lawyers who understand how to deploy a Skill for the purposes of legal research (or drafting or whatever) will at least be showing a level of competence that has sometimes been missing within the profession. Who knows, maybe Skills are the missing piece that will improve professional confidence in using AI. At the very least, like figuring out how your web browser works and being able to install an extension, or how to style a Word document and run a Table of Authorities, Skills seem to be one of the things that would be easy for law students to be literate about as the technology develops during their law school years.