General-purpose legal AI has a ceiling. It can explain a concept, summarise a document, or draft a simple letter. But when the work demands the logic of a specific practice area (the deal structure of an M&A transaction, the evidentiary chain of a criminal defence, the regulatory map of a compliance review), the model needs more than a prompt. It needs a skill.
What a skill contains
In CouncilGPT, a skill is a domain module loaded on demand by the Council. It includes the frameworks that govern analysis in that practice area, the types of authorities that matter, the red flags to surface, and the citation conventions to follow. A corporate skill knows how to read a shareholders' agreement for change-of-control triggers. A litigation skill knows how to assess whether a clause supports a motion to dismiss. A compliance skill maps facts to regulatory obligations.
Skills are not hardcoded casebooks. They are living instructions that the legal knowledge team updates as the law changes and as customers refine their needs. When a lawyer asks a question, the Router Agent selects the appropriate skill before the Council begins its work.
The current catalogue
Today the platform ships with skills for corporate, litigation, compliance, immigration, personal injury, and forensics work. Each skill is built with practitioners and tested against the kinds of questions those practitioners actually ask. The goal is not to replace a specialist; it is to give every lawyer a specialist's lens on demand, grounded in their own documents and instructions.
Specialisation without silos
Because skills are composable, a cross-border M&A question can load both corporate and compliance skills, and a product liability matter can load litigation and forensics skills together. The Council coordinates them so the final answer reflects the full picture. Practice-area depth, available wherever the question lives.