Agent Context Layer

Structure the documentation and context of a code repository so an agent can build an accurate mental model of the project fast, then load only what the current task needs. This is the Interpretable Context Methodology (ICM) by Jake Van Clief (also published as the Model Workspace Protocol; source of record: ~/Developer/AI/Standards/ICM, _core/CONVENTIONS.md) applied to the context layer of a code repo.

Think of the repo's docs as a library the agent walks. The entry file is the catalogue: small, stable, it points at everything and stores almost nothing. Content lives on the shelves. The agent starts at the catalogue, follows one or two pointers, and stops as soon as it has enough.

Scope

This skill owns the context/documentation layer only. It does not turn a code repo into a staged pipeline.