Six primitives, one coherent system
Copilot's customisation stack works in layers: instructions are always on, skills load when relevant, prompts are invoked manually, custom agents serve dual roles (persona agents for interactive sessions and task agents that run autonomously in chains), and hooks enforce hard policies.
A governance layer — asset manifest, model compatibility matrix, CI evaluation gates, and changelog — ensures quality and ownership as the setup evolves.
Every file below has a model: field in its frontmatter — this routes each task to the AI model best suited for it automatically, without you switching the picker. Click any row to see the full rationale.