Exploded axonometric · Diagram Design

AI agent stack · What sits under the prompt

Five layers pulled apart along one axis. You type at the top, the harness runs the model, and skills decide how the work gets done.

AI agent stack · What sits under the prompt Exploded view of an AI agent stack: a secrets vault at the base, then tools, skills, the agent harness that runs the model, and the chat or terminal interface on top, with skills as the focal layer. Secrets vaultscoped service token Toolsclis, apis, mcp, scripts Skillsmarkdown playbooks Agent harnessclaude code, codex Interfacechat or terminal FOCAL LAYER Skills are plain files the agent reads before it acts. Swap the harness or the model and they still work.

The headline

Skills are the layer you own

The interface and the harness come from a vendor and change every few months. A skill is a markdown playbook in your repo, so it carries over when the rest of the stack is swapped.

Reading the explode

  • Top layer is where you type
  • The dark chip is the model inside the harness
  • Each card on the skills layer is one playbook
  • Plugs on the tools layer are the CLIs and APIs it can call

Why the vault is at the bottom

Every tool call runs on a scoped service token. Keep it under everything else so no layer above holds a raw secret.

ai agent stack · exploded axonometric example · diagram design