Antoine Louis’ post is worth reading because it captures an important shift in legal AI: Lawve AI is moving legal "skills" from isolated prompt files toward a shared infrastructure layer for professional method.
That inspired me to contribute a Swiss Legal Source & Authority Triage skill to the Lawve AI / awesome-legal-skills repository.
The idea is simple:
Before an assistant answers a Swiss legal question, it should first identify which authority layer governs the issue.
Federal law, cantonal law, communal rules, regulator practice, registers, doctrine, soft law, contracts, professional standards, multilingual official versions - these are not interchangeable sources.
Swiss legal AI does not only fail when it gives the wrong rule.
It also fails when it answers from the wrong layer of authority.
My contribution is intentionally foundational. It does not try to turn an assistant into a Swiss lawyer. It gives the assistant a method for source mapping, source-status classification, official-source checks, uncertainty flags, and escalation to qualified Swiss legal review where needed.
The governing principle is:
Route authority before generating an answer.
If legal AI is going to become useful in real legal work, we need more than better outputs. We need reusable professional methods that make authority, proof, review boundaries, and source status explicit.