AI Learning Hub › AI Skills
For Stanford Law students
Twenty-one AI skills built for law study. Each one is problem-first, policy-aware, and organised around verification. Take a whole set in one download or a single skill on its own, then install it in ChatGPT or Claude — you add the ZIP once and it stays available across chats.
New · case study
Before you install one, see how a skill is actually built. We take apart Anthropic’s
open-source Claude for Legal project — what a SKILL.md file is, why the
work is split into small skills, and why an open skill does not make Harvey or Legora obsolete.
About 25 minutes · no coding background assumed.
Or listen · audio explainer
21 min 55 sec The same ground as the written case study, spoken. Open in Google Drive (opens in a new tab)
One download
Take a whole set at once. A set is a single ZIP holding the skill files it names: unzip the set, then upload each skill ZIP without unzipping it. Every skill here works in both ChatGPT and Claude, and every one can also be downloaded on its own below.
Ten skills that review your writing without writing it for you: the SLS policy gate, the full review workflow, and eight focused checks you can run on their own. Every one of them flags, explains, and asks — none of them drafts, rewrites, or replaces your prose.
The five foundational skills in order, from your first AI orientation to auditing AI output like a lawyer.
Every guided studio in one file — Harvey, Legora, LexText, CICERO, Gemini Notebook, and the explorer for anything not on the list.
Start here · in order
Five foundational skills, from your first AI orientation to auditing AI output like a lawyer.
Start here: how AI actually works, the SLS student AI policy, the PAUSE Rule, and your first learning-and-disclosure record.
Decide whether AI is permitted, appropriate, and the right fit for a task — and get an AI Learning Plan with a clear go / no-go.
A Socratic partner for reading and briefing opinions. You attempt first; it questions, points to the text, and never writes your brief.
A traditional-research-first workflow: validate authorities in authoritative sources, then use permitted AI to compare and stress-test.
Audit AI output claim by claim — checking whether authorities exist, are current, and actually support the point.
For a draft you have written
Ten skills for a draft you have already written. Each one reviews as a supervisor would: it identifies the problem, explains why it matters, points to the exact place, and hands the revision back to you. None of them drafts, rewrites, or supplies replacement prose. Start with the gate, then either run the full review or pick the check you need.
Run this first. Works out whether AI review of this draft is authorised under your course policy or the SLS default — and flags confidentiality and professional-responsibility problems before anything reads the paper.
The whole workflow in one skill: the policy gate, then argument, organization, clarity, audience, counterarguments, sources, genre, and Bluebook — and a Word review copy you can work through.
Tests the thesis, rule synthesis, evidence chain, and conclusion for gaps, then asks the revision questions instead of answering them.
Whether the draft is ordered, signposted, and followable: sequencing, headings, transitions, repetition, and points buried where no reader will find them.
Ambiguity, jargon, dense sentences, and inconsistent defined terms, flagged where they sit — with revision strategies you carry out yourself.
How a professor, judge, supervising lawyer, client, editor, or employer is likely to receive the draft: tone, credibility, framing, and unstated assumptions.
The strongest objections, missing adverse authority, and doctrinal soft spots — issue-spotting and questions only, never a rebuttal written for you.
Traces each claim to its source and separates “this source exists” from “this source supports the point,” ending in a claim-source matrix.
Bluebook 22nd-edition issue-spotting: short forms, pincites, signals, order, and typography. It names the likely defect and the rule; it does not write the citation.
Whether the draft behaves like the form it claims to be — memo, brief, seminar paper, policy paper, client letter, statement — on purpose, stance, sections, and conventions.
Something to practise on
Five fictional student drafts, so you can watch a review skill work before you hand it your own writing. Each one has real problems in it — that is the point. Download a draft, install a skill, upload the file, and ask for the review.
Everything in these files is invented. The parties, the facts, the drafting, and the student authors are fictional, and the documents were written as teaching samples. Do not rely on any authority cited in them, and do not treat any of them as a model answer.
A one-page brief of a negligence appeal: facts, procedural history, issue, rule, holding, reasoning, takeaway. The rule statement claims more than the case decides.
Try it withGenre Fit · Argument and Structure · Clarity and Precision
A short first-year memo with three headed elements and two authorities. The register slips towards the conversational, and the brief answer is more confident than the analysis under it.
Try it withClarity and Precision · Audience and Reception · Bluebook Audit
A longer memo working through the bystander elements one at a time, conceding the easy one and spending its argument on the contested one.
Try it withArgument and Structure · Counterargument · Flow and Organization
The most developed of the drafts, and the most heavily cited: six authorities, a quoted email, and a conclusion that has to weigh them.
Try it withClaims and Source Traceability · Bluebook Audit · Counterargument
A timed answer on negligence and a rescuer, written under exam conditions: elements marched through in order, no citations, conclusions ahead of the analysis.
Try it withArgument and Structure · Flow and Organization · Audience and Reception
For a class, a study group, or yourself
A session built from what is on this page: set-up in ChatGPT and Claude, five activities to choose from, discussion topics with what to listen for, an explainer note on every skill, and a glossary of the jargon. Faculty can run it as it stands; on your own it is a structured way through the whole set.
60 or 90 minutes · or at your own pace.
Explore when ready
Guided, Stanford-specific introductions to the AI tools available through the Robert Crown Law Library and Stanford.
A problem-first, SLS-specific introduction to Harvey AI, with bounded practice and source verification.
Map your workflow, see where Legora helps, and practise on safe documents while verifying every source.
Decide whether LexText fits a task, then learn a current feature through focused, safe practice.
Prep an argument map and goals, run the simulation in CICERO, and debrief into a focused improvement plan.
Source-grounded study with a curated set of readings — inspect every cited passage instead of trusting it. Formerly NotebookLM.
Bring any AI tool or feature and get a current, guided tutorial that starts from your problem.
What every skill has in common
The skills take a problem-first approach and build in the PAUSE Rule, Stanford Law School student AI guidance, Responsible AI at Stanford, productive struggle, legal-source verification, structured and manageable data, transparent AI-use logs, and accurate, non-anthropomorphic explanations of how AI systems work.
Check your course policy first. Confirm your course’s AI policy with the instructor, and remember the Honor Code applies either way. Use Stanford’s own accounts — ChatGPT Edu (opens in a new tab) and Claude (opens in a new tab) — rather than a personal one. Access to Law Library–provided AI tools is limited to current SLS community members and uses Stanford sign-in. For access, tools, or research help, email library@law.stanford.edu.