We strongly recommend to utilize an AI model to create a codemod for your changes, such as Cursor with claude-4-sonnet.
Here is a list of instructions that will help the AI model to come up with a better result
- Start all input/output fixtures files with `// @ts-nocheck`. Make sure the comment remains in place in the output fixture file.
- Update `packages/codemod/src/lib/upgrade.ts`
- Use `import { createTransformer } from './lib/create-transformer';` for codemods. Do not import anything from `jscodeshift` directly.
- No need to cover imports that use `require()`
- The codemod should not return anything. It should set `context.hasChanges` to `true` instead.
- See files in `packages/codemod/src/codemods` for conventions
- Multiple input/output files can be used in case of import conflicts.
- Run tests to verify the change
- Run the codemod manually to verify that it's working
- If you need to create temporary files for testing, create them in `packages/codemod/`, and remove them when done.Depending on the complexity of the changes, you can instruct the AI to review changes directly from a pull request, e.g. https://github.com/vercel/ai/pull/5750.diff. If that doesn't yield a useful result, try describing the breaking change such as in the example below
# Breaking change
## `streamtext()`: `result.file.{mediaType,data}` properties is now `result.{mediaType,data}`
Before:
```ts
import { streamText } from 'ai';
const result = await streamText({
model: someModel,
prompt: 'Generate an image',
});
for await (const delta of result.fullStream) {
switch (delta.type) {
case 'file': {
console.log('Media type:', delta.file.mediaType);
console.log('File data:', delta.file.data);
break;
}
}
}
```
After:
```ts
import { streamText } from 'ai';
const result = await streamText({
model: someModel,
prompt: 'Generate an image',
});
for await (const delta of result.fullStream) {
switch (delta.type) {
case 'file': {
console.log('Media type:', delta.mediaType);
console.log('File data:', delta.data);
break;
}
}
}
```