Every entry here is a written specification you paste into an AI assistant that can connect to your systems. The assistant interviews you, configures itself, runs once while you watch, and saves. No vendor, no seat license, no integration project.
1. Pick an entry and copy the spec. 2. Paste it into an assistant that can connect to your tools. 3. Answer its setup questions honestly — the specificity is what makes it useful. 4. Watch the first supervised run and correct it. 5. Save it, and only then let it run on a schedule.
Nothing sends, files, publishes, or touches a client without a named human approving it first. If you edit a spec, that is the line to keep. An agent that drafts is a tool. An agent that sends on your behalf is unsupervised nonlawyer assistance, and that is your problem, not the vendor's.
A spec tells an assistant what to do. It does not tell you how to choose a model, where your data goes, what to write into an engagement letter, or how to supervise a system nobody at the firm built. Jacob Malherbe wrote A Lawyer's Guide to Building AI on exactly that — the setup work that sits underneath every entry in this index.
Specs are reviewed by hand before publishing. Anything that automates client contact without approval, generates review text, or contacts represented parties gets rejected. Add yours.