What are 'skills' in the Stack AI tool and how do they help customize the agent's behavior for different clients?

Skills are described as large standard operating procedure documents that give Stack an understanding of how a firm or user likes to work. They live at the console level and can apply across all clients (the presenter mentioned 55 clients using a default skill, with some clients having customized variants). Skills are invoked automatically by the agent based on the prompts given — for example, a variance analysis prompt automatically invoked a variance analysis skill. After completing a task, the tool can be asked whether it learned anything that should update a skill; in the demo, it identified that class should be included on entries going forward and drafted an update to the 'schedules and accrual' skill, which the user then approved and saved. Ramp provides default skills as a starting point seeded into the advisor console and client-level environments, and from there users can create more custom skills or have Stack update the defaults. The presenter advised that when starting to use Stack, users should not feel they need to build out all these custom skills before getting started — this is considered a more advanced use of the product.