Before you start
You need Node.js 20 or newer (
node --version) and a Ditto account (app.dittowords.com)../src, ./app, or the repo root on a small project.
Run the scan
- Claude Code
- Cursor
- Codex
- Terminal
Paste this prompt into Claude Code from the root of your repo.
This only needs read access to your codebase. It won’t edit or write to your repo. You’ll be able to review what’s scanned before bringing it into Ditto.
Review and refine
Ditto classifies the strings and shows how many strings and locales it found, then takes you through setting up your content system:- Set up your content system in Ditto. Ditto shows the unique and repeated strings it kept, the developer IDs it found, and your product name and blurb. Mark any strings to leave out or handle differently.
- Review proposed content system. The tags Ditto will create, how it will give components names, how it will organize them, the variants and locales it found, and how it will name developer IDs, each with examples. Describe what you’d like to adjust, in plain language, until it looks right.
- Create. Ditto creates the components, variables, and 1 variant per locale. If your library isn’t empty, they’re added in a new top-level folder named after the scanned folder. Home shows what was imported, and you can search and reuse what’s live today across your product.
Check your text
While Ditto builds your content system, it also checks the scanned text for common text issues. The results appear on the review page under Here’s what we found, with a count for each check:
Select a check, or Review all, to see each finding and every place it appears in your code. The checks never change your text, and nothing they find blocks setup.
Flag the findings you want to fix, and Dismiss the ones that are fine. When a finding has more than one version, like “Sign in” and “Log in”, you choose which versions to flag. When your library is created, components with flagged text get the In Review status, so they’re easy to find and fix. Dismissed and unreviewed findings leave your text as it is.
Only text Ditto is confident your users see is checked. Spelling, grammar, wording, capitalization, and punctuation checks cover English text; translation keys are caught in every locale.
Get a first style guide
We can infer a starter style guide from your strings, or you can import your own. After a scan finishes, Ditto drafts a style guide with three sections: Voice & Tone and Style & Mechanics (rules) and a Word List (terminology). Each rule has a name, a description, and up to three from → to examples taken from your copy. Definitions: Style guides and rules.- In Ditto, open Style guides in the left sidebar. The inferred guide appears as a draft named after your product (or “Content Style Guide” if the name could not be inferred).
- Open it. Edit, delete, or add rules, and move rules between sections.
- Enable it. Turn on Enabled by default so it applies to every project, or enable it per project from the project’s … menu under Style guides.
get_styleguide_rules as soon as the guide is enabled; the PR review bot checks pull requests against them. To add, refine, or retire rules from your editor, see Agent instructions.
Frequently asked questions
What if my repo already has translation files? Scan readsstrings.xml, .strings, .stringsdict, .xcstrings, and JSON catalogs, and each locale becomes a variant in Ditto with that locale’s text.
Can I scan a workspace that already has components? Yes. New components are added in their own top-level folder (for example, src). Scanned text isn’t matched to existing components, so duplicates are possible.
What happens if I scan the same repo again? Create adds a new, unlinked copy of every component (for example, in src or src 1). To keep only the latest import, delete the earlier components. Before deleting them, check their developer IDs and update any integrations that use them. The new components can have different developer IDs.
Can I run the text checks without importing? Yes. Review the findings, then leave without selecting Create. Nothing is added to your library.
Is there a limit? Trial workspaces can classify up to 100,000 strings in total across all of their scans. Paid plans have no cap. If a scan would exceed the remaining budget, the CLI stops before uploading and suggests subdirectories that fit.