# Agent Learnability Surface

Capsulang is broad because it is an agentic runtime, not only a systems language. To keep it learnable on demand, the reference implementation exposes a small agent-facing path:

```text
pattern -> .caps module -> check -> test -> graph -> cost -> release preview
```

## Commands

```bash
caps patterns list --json
caps patterns show regulated-deployment-approval --json
caps new access-change-approval --out access_change.caps

caps check access_change.caps --json
caps test access_change.caps --json
caps agent-graph --json access_change.caps
caps cost --json access_change.caps
caps repair-plan --json access_change.caps
caps release-preview --json access_change.caps
```

## Teaching agents

Use the bundled Codex/Claude support files:

```bash
scripts/install_agent_support.sh --repo
scripts/install_agent_support.sh --user
caps skills list --json
caps skills get capsulang-builder --platform codex --full
```

## Default design rule

Agents should not invent an approval workflow from scratch when a standard pattern exists. Start from `caps patterns list`, instantiate with `caps new`, then customize.