When does Kwern reach for localization?

Write an app Product's App Store metadata for its priority storefronts — title, subtitle, keywords field, description per locale — and choose which storefronts deserve it. That sentence is the skill's own description, which is what a model reads to decide whether to use it; the rest of the file is on this page.

Which Kwern agents run localization?

The ASO agent can activate it during a run, when what it finds calls for it.

Do I have to sign up to read the file?

No. The whole file is on this page, and the raw Markdown is at https://kwern.app/skills/localization.md. Nothing here is behind a sign-up.

Can I use localization outside Kwern?

Yes, and you should. Kwern wrote localization and publishes it exactly as it runs it. It expects Kwern's own tools, the Product and its connectors and its memory, so outside Kwern you get the method without the data path under it.

What does Kwern add that the file alone does not?

The data and the cadence. A skill file is a method; Kwern runs it against your foundation doc and your connected accounts, on a fixed cadence, and leaves what it wrote in an approval feed. The file is the method, the agent is the method with your numbers in it.

localization

Write an app Product's App Store metadata for its priority storefronts — title, subtitle, keywords field, description per locale — and choose which storefronts deserve it.

Written by Kwern, after the skill of the same name in Eronred's aso-skills, MIT. Updated 2026-10-05.

The file, as Kwern runs it

Nothing below is rewritten or cut. The description a model routes on comes first, then the file itself. The whole thing, frontmatter included, is at /skills/localization.md.

description

Write an app Product's App Store metadata for its priority storefronts — title, subtitle, keywords field, description per locale — and choose which storefronts deserve it. Use for "localize my listing", "which countries should I target", "translate my metadata", "expand to Germany/Japan". For the primary storefront use metadata-optimization.

Bring the Product's listing to the storefronts where visitors already arrive and convert least. Output is one listing Proposal per field and Locale, which Kwern writes to App Store Connect when the User approves it, plus a per-storefront metadata package with the reasoning, saved with save_output. Keywords in another language are researched, never translated: "budget tracker" is "Haushaltsbuch" in German and "家計簿" in Japanese, two different concepts.

Gather

read_context — markets the Foundation doc names, whether the app UI is localized and in which languages, voice, competitors.
connector_asc_analytics when the ASC Connector exists — impressions, product page views, downloads and conversion per territory, 30 days plus trend. Without it, take the markets from the Foundation doc; without those, start from the tier-1 list in the references.
connector_asc_localizations when the ASC Connector exists — every locale already localized, primary first, with its title, subtitle, keywords field, description and promotional text; a locale whose text repeats another locale's is a copy, not a localization. Without it, iTunes Lookup through fetch_url (https://itunes.apple.com/lookup?id=<App Store id>&country=<storefront>) returns the name and description visitors of that storefront read, and the subtitle and the keywords field stay unknown; say so in the package.
connector_apple_ads_search_terms when the Apple Ads Connector runs campaigns in the storefront — the terms locals search and convert on. Without it, infer from local competitors and reviews.
The competitors in each priority storefront, through fetch_url on iTunes Search (https://itunes.apple.com/search?term=<name>&entity=software&country=<storefront>&limit=1) — how they phrase their name and description in that language.
app_store_reviews per storefront — the vocabulary users write in, and a quotable line per language.

Choose the storefronts

Score each candidate on the territory data: high impressions with low conversion is the strongest signal (visitors arrive, the listing loses them); then downloads trend, whether the app UI speaks the language, and how many competitors are localized there. Take the top three to five; state the score per storefront. The locale codes, the storefronts each locale covers, and the tier lists are in references/storefronts.md — read it before naming a locale, because one locale serves several countries and the default locale serves every storefront without its own.

Write per storefront

Field limits, indexing, formulas and the mistakes to flag are in ../metadata-optimization/references/field-rules.md; they hold in every locale, and character limits count characters, so CJK and Cyrillic fit fewer words. Cultural checks for description, screenshots and pricing are in references/cultural-adaptation.md — read it when the storefront differs in script, formality or currency from the primary one.

Per storefront, in this order:

Keyword set — five to ten local terms with the search intent behind each, drawn from search terms, competitors and reviews; the English equivalent beside each so the User can verify with a native speaker.
Title and subtitle — brand plus the local #1 keyword, benefit-led subtitle, formulas from the field rules; a translated tagline that carries no searched term is replaced.
Keywords field — built from the local set, singular forms, zero words repeated from title and subtitle.
Description — adapted, with the hook rewritten for the local framing, local social proof when it exists, and a review quote in the language.
Promotional text — for the local moment (holiday, season, event).

Every field arrives with a back-translation line so the User can read it.

Output: per-storefront metadata package

For each priority storefront: locale code, the storefronts it covers, the score and its reason, then every field with chars/limit, keywords covered, back-translation, and a one-line rationale. Then:

Priority matrix — territory × impressions, conversion, trend, UI localized, competitor localization, score.
Before / after — the current localization per field when connector_asc_localizations or iTunes Lookup returned one.
Creative to-do — the captions and pricing screens screenshot-optimization and app-preview-video must re-render per storefront.
Apply — which fields need a version the User can still edit in App Store Connect: title, subtitle, keywords field and description do; the promotional text does not.

Propose

Each field you write is one propose per Locale, with appStoreField (name for the title, subtitle, keywords, description, promotionalText) and the locale code in locale: content is the field in that language, whole, and why names the local keywords it covers and the territory signal behind the storefront. A Locale the listing lacks is added by the first Proposal approved for it.

Done when every field of every priority storefront has its Proposal, each storefront has every field within its limit, every local keyword appears once across title, subtitle and keywords field, every field carries a back-translation, and the package names the data it lacked (territory analytics, keywords field, search terms) when a Connector was absent.

metadata-optimization — the primary-storefront package this Skill mirrors.
screenshot-optimization — captions per storefront.
app-preview-video — preview captions and re-recording per storefront.
seasonal-aso — local holidays that drive the promotional text.

What it discloses

The Skill reads these files only when it needs them, so the model spends nothing on them the rest of the time.

references/cultural-adaptation.md
references/storefronts.md

From your accounts to a proposal

01
Connect what it reads
Give the ASO agent the accounts it works on. Kwern never asks for a password, and an agent missing a connection says so instead of guessing.
02
The loop runs weekly
Before it writes a word it reads the foundation doc, which holds the positioning, the audience and the voice, and whatever its connectors return that day.
03
The localization skill runs
The agent activates it when the run calls for it, and what the model then reads is the file on this page, word for word.
04
A proposal waits for you
Nothing is published and nothing is spent until you approve it. Approve and Kwern posts it where it holds the keys; where it does not, it hands you the text and the link.

The file, the agency, and the agent

	On your own	An agency	Kwern
The method	Whatever you remember to do that week	Theirs, and you never get to read it	This file, open on this page
Who runs it	You, when there is time	An account manager, between other accounts	The ASO agent, weekly
What it reads	What you can pull by hand	What you remembered to send them	Your connected accounts and your foundation doc
What comes out	A draft, sometimes	A deck, and a report at the end of the month	A proposal you approve or throw away
What it costs	Your evenings	A retainer	One subscription, and the model on your own key

Questions about localization

Suite: kwernskills
Author: Kwern
Updated: 2026-10-05

Install it in Claude Code:

# Claude Code, in the project you want it in
mkdir -p .claude/skills/localization && curl -o .claude/skills/localization/SKILL.md https://kwern.app/skills/localization.md

# or just read it
curl https://kwern.app/skills/localization.md

Raw SKILL.md