## Description This issue was discovered on a project where `field_prefix` in `field_ui.settings.yml` was set empty, causing all custom fields to be silently skipped during AI translation even though they were translatable and had registered extractor plugins. Turns out `TextExtractor::shouldExtract()` uses a hardcoded list of field-name patterns to decide which fields are extracted for translation: https://git.drupalcode.org/project/ai/-/blob/1.3.4/modules/ai_translate/src/TextExtractor.php?ref_type=tags#L133-139 ```php // @todo better way to find fields that should be translatable? static $supportedFieldNames = ['body', 'info', '^field_', '^layout_builder_']; ``` The `field_` prefix is a convention, not a requirement — it can be changed or omitted entirely via `field_ui.settings.yml` (`field_prefix: field_`). ## Tasks * [ ] Replace the hardcoded name-pattern check in `TextExtractor::shouldExtract()` with an `instanceof FieldConfigInterface` check, deferring to the plugin's own `shouldExtract()` logic for all custom configurable fields. ## Acceptance criteria * A translatable node field whose machine name does not start with `field_` is included in AI translation output. ## Testing instructions 1. Set `field_prefix` to empty in `field_ui.settings.yml`. 2. Create a plain text field named `summary` on an Article node. 3. Run AI translation on a populated Article node. 4. Verify the `summary` field content is translated. ## Related issues