I'm a Texas appellate attorney building and testing AI workflows for high-stakes legal writing — a domain where a hallucinated citation is not a quirk but a sanctionable event.
Everything in this repository was built on my own dime, outside billable hours, and is free to copy, adapt, and use. No attribution required.
Verity — the citation-verification pipeline (skill slug: cite-check). A multi-round process for auditing AI-assisted filings against ground-truth judicial opinions. Round 1 catches nonexistent cases. Round 2 verifies every quotation against the actual opinion. Round 3 checks parentheticals and characterizations of holdings. Designed to run on a real brief in minutes, with a human-approval gate between rounds.
Benchmarked prompt library. Reusable, genericized prompts for appellate workflows — record analysis, issue-spotting, outline building, drafting — each documented with where the model is reliable and where it fails.
CLE curriculum (in development). Converting these findings into accredited continuing legal education, so the work scales from one practice to the profession.
| Prompt | What it's for |
|---|---|
| The Legislation Drafter | Article V for laypeople. Turns a problem someone already understands into a properly formed bill, a plain-English vet memo, and a list of questions for a lawyer. Never gives legal advice. |
| Motion for Rehearing → Petition for Review | Converting error-correction into review-worthiness. A conversion table maps each rehearing argument to a review-granting factor — or recommends dropping it — before a word of the petition is drafted. |
| Oral Argument Preparation | A paired set: a comprehensive moot question bank weighted to financial exposure and aimed at your own side's weak points, and a short visual podium primer condensed from it — so the advocate and the moot panel are working from the same answers. |
| Reporting Harassment at Work | Builds the dated log and the report, in your own words. Distinguishes I don't remember from I remember and there's no paper — because those are opposite facts and only one of them is a weakness. Pause anytime; the record keeps your place. |
| The College Application Keeper | For a student applying without a counselor, a parent who has done it before, or money for either. Most of what a private counselor does is bookkeeping — deadlines, fee waivers, aid forms, and a record of what you have actually done. That part can be given away. |
| The Steady Companion | Guardrails for AI in mental-health conversations — a standing prompt you write on a calm day, so the model meets you instead of managing you, and keeps your care team in the room instead of replacing them. |
| The Genie Lamp | A starter kit for persistent memory. A model has no memory between sessions, so the state lives in files you re-inject at the top of each one. Includes the folder, the starting contents of every file, and the setup prompt. |
| The Beginner Explorer's Pack | A starter continuity kit for anyone new to working with an AI across more than one session. Installs three habits — a ledger, versioned autosaves, and a handoff — so the work doesn't vanish when the session ends. |
| The Prompt-Library Keeper | The meta-prompt. Turns any good working prompt into a durable library entry and keeps the library consistent as it grows. Chat logs are disposable; the library is the real memory. |
| "I Play Better When I'm Drunk" | State-dependent memory for AI-assisted work. A one-page grip card that reconstructs how a long document was made before asking you to defend what it says. |
| My Genie Is Not White | A standing frame-check and lessons ledger for builder-default bias — the defaults of whoever built a system becoming invisible walls inside it. |
Verification before launch. I come from a family of systems-builders (Aegis, mathematical modeling); this is the language version of that discipline. The question isn't whether lawyers will use AI — it's whether they'll use it with guardrails or without.
First prompt: January 2026. This repository documents what came after.