attac-t Skill: Consumer First
"Write the call-site you wish existed. Then make it true."
The Standard
- Consumption before implementation: Before building a domain, write the code that will call it. If it doesn't read well, the shape is wrong — fix it while there is nothing to refactor.
- Less code is better code, with one exception: Every internal line is a liability. A line that makes a call-site clearer is an asset.
- The call-site is the product: Your domain has one consumer before it has a hundred. Design for them, not for the implementation.
- Return the thing, not a field of it:
owner(), neverownerId(). The caller who wants the key writes->id; the caller who wants anything else is not sent back for a second method.
The Check
Of every line you are tempted to keep, ask:
Which call-site becomes clearer because this line exists?
- An answer → DX surface. Keep it. Named predicates, query scopes, expressive fluent methods.
- No answer → liability. Delete it. Defensive guards, speculative parameters, pass-through wrappers, narrating comments.
This is why extracting a condition into a named predicate is correct even though it adds lines. The if now reads as a sentence. The line was bought, not spent.
The Anti-Patterns
| ❌ Don't | ✅ Do | Why |
|---|---|---|
| Build the domain, then find the call-site ugly | Sketch the call-site first, then satisfy it | Reshaping is expensive; sketching is free |
| Delete a named predicate to "save a line" | Keep it — the caller reads as a sentence | It is DX surface, not internal weight |
| Add a parameter "in case someone needs it" | Add it when a call-site needs it | No caller is clearer for it today |
| Wrap a method to "keep the API consistent" | Let the caller use the real method | A pass-through answers the question with silence |
| Return an id from a resolver | Return the model; the caller takes ->id |
One question answered, the rest hidden |
Real-World Examples
See examples.md.