| name | domain-model-context |
| description | Load a domain model into the working context before a coding task. Use at the start of a session that will touch product concepts, when the user says "use the domain model" or "load the model," or before implementing a feature in a repo that has a *.modelith.yaml. Makes the agent reason in the team's vocabulary — entity names, relationships, and invariants.
|
Loading the domain model as context
A well-formed domain model is a type system for the problem space. Loading it
before a coding task means you use the team's exact names, respect the declared
relationships, and don't violate invariants.
What to load
- Find the domain model in the repo — a
*.modelith.yaml and its rendered
*.modelith.md (often under docs/ or the repo root).
- Prefer the rendered Markdown for reading: it's the readable form, with
the relationship diagram inline. If it's missing or stale (check with
modelith render --check <file> — the file argument is required), regenerate it with
modelith render <file> first. (If modelith isn't installed, you can still read the
committed .md directly; just note it may be stale.)
- Read it in full before writing code. Internalize:
- the canonical names — use
Project, never "workspace" or "container";
- the relationships and cardinality — what owns what;
- the invariants — rules your code must not break.
- If it has an
imports: list, its scope.Name references (in attribute
type values) name vocabulary that lives in the imported file, not this
one — load that file too before treating a scope.Name term as unknown.
How to apply it while coding
- Name variables, types, functions, and UI strings using the model's terms.
- When a requirement seems to need a concept the model doesn't have, stop: that
may be a real gap. Flag it and offer to capture it with the
domain-model-author skill rather than silently inventing a name.
- When code would violate an invariant, treat that as a bug in the plan, not a
detail to smooth over.
Keep the model honest
If implementing the feature reveals the model is wrong or incomplete — a missing
entity, an invariant that can't hold — surface it. The model is meant to be a
living source of truth; coding against it is exactly when its gaps show up.
