Coding

Code generation

Turn specs and requirements into working code.

Knowledge cards
Knowledge card
Generate a Full Module From a Written Spec
Instead of asking Claude for a single function, you give it a short written specification — inputs, outputs, constraints, and file layout — and it returns a complete, coherent module.
L3 · Workflows
Workflow
Generate One Layer at a Time, Not the Whole Feature
A workflow where you ask Claude to build a feature one layer or slice at a time — data model, then business logic, then the endpoint, then the UI — verifying each before moving on, rather than requesting the entire feature in a single generation.
L3 · Workflows
Best practice
Give Claude the Types Before the Logic
A code-generation pattern where you (or Claude, with your approval) fix the types and function signatures first — the data shapes, inputs, and return types — and only then ask for the implementation that satisfies them.
L2 · Context engineering
Knowledge card
Point Claude at a File to Copy the Style
A code-generation pattern where, instead of describing your conventions, you hand Claude one existing file that already does something similar and ask it to write the new code in the same shape.
L2 · Context engineering
Claude code generation

Give Claude the types first

Claude writes noticeably better code when it knows the shapes it is working with. Give it the types, the function signature, and the interfaces before asking for an implementation, and the output slots into your codebase instead of inventing its own data model. This is especially true in typed languages, where the types are a precise spec Claude can honour. Handing it the contract first turns code generation from a guess into a fill-in-the-implementation task, which is where it is strongest.

Generate one layer at a time

A whole feature in one response is hard to trust and harder to review. Generate in layers — the model, then the logic, then the interface — and check each before moving on. Give Claude the surrounding context so each layer fits the last, and describe the constraints it must respect. For anything non-trivial, this staged approach catches wrong assumptions early, when they are one function to fix rather than a tangled feature to unpick.

Frequently asked
How do I get better code from Claude?

Give it the types, signatures, and interfaces before asking for an implementation, so it honours your data shapes instead of inventing its own. The contract first makes the output fit your codebase.

Should Claude generate a whole feature at once?

No — generate one layer at a time (model, then logic, then interface) and verify each. Staged generation catches wrong assumptions while they are still small and easy to fix.

Why does Claude's code not match my codebase?

Usually because it lacked the types and surrounding context, so it invented a data model and conventions. Provide those up front and the code fits far more often.