name	translate-missing
description	Translate `TODO translate`-marked keys from generated locale files into the corresponding `.template.` files. Use when the user asks to fill in missing translations, finish translating a language, or process TODO translate markers.
user-invocable	true
allowed-tools	["Read","Edit","Write","Grep","Glob","Bash"]

/translate-missing

Translate keys marked /* TODO translate; non-english translations: ... */ from the generated locale files into the matching *.template.* files.

Background

pnpm sync-language-files (driven by translation-manager/sync-language-files.ts and translation-manager/templates.json) regenerates the plain locale files from en plus each *.template.* file. Any key that is in en but missing from a <lang>.template.* gets emitted into the generated file with a TODO translate block comment that includes the English value and any sibling-language translations as hints. The *.template.* files are the source of truth; the plain files are gitignored and overwritten on each sync.

The set of governed files is not fixed — read it from translation-manager/templates.json (.templates[], each carrying a {LANG} placeholder) so this skill stays correct as feature folders are added. The English source is {file}-en.{ext} / {LANG}=en (no .template); it has every key and is what the templates fall back to via DeepPartialWithRequiredObjects<Messages>.

Which languages

Do not hardcode the language list — it grows. The canonical codes are the languages tuple in src/shared/langUtils.ts, re-exported through src/features/mainMenu/components/languageItems.ts; en is the master and is never a translation target. The sync iterates exactly translation-manager/templates.json's .langs[], which must match. List the current targets with:

jq -r '.langs[]' translation-manager/templates.json

If $ARGUMENTS names specific languages, process only those; otherwise process every code from the command above.

Steps

Run pnpm sync-language-files to make sure the TODO markers reflect the current state.
For each language, expand its generated files from templates.json and scan them for TODO blocks. Generated *.tsx / *.ts files are gitignored, so the Grep tool won't see them — go through Bash:
```
# every generated file for <lang> that still has TODO markers, with counts
jq -r '.templates[]' translation-manager/templates.json \
  | sed 's/{LANG}/<lang>/' \
  | xargs grep -c "TODO translate" 2>/dev/null \
  | grep -v ':0$'
```
Blocks look like:
```
/* TODO translate; non-english translations:
 * sk: 'Predvolený'
 * cs: 'Výchozí'
 */
default: 'Route default',
```
The block carries everything needed: the English value (the last line) and any sibling translations. Translate into <lang> using both as context — don't blindly copy the Slovak.
For each TODO key, add the translation to the matching *.template.* file at the same path inside the messages tree:
- Locate the parent object in the template by searching for sibling keys that already exist (the English source shows which siblings to look for).
- Use the shape-spread / object-literal style already in the file. Add the new key alongside existing siblings.
- If a parent object exists as {} or is missing entirely, fill it in / create it. Some sections (e.g. gallery.stats, wikimediaCommons) are empty placeholders in some templates.
- If a top-level section is missing (e.g. access, emergency, entrance in OSM mappings), add it where sibling sections sit.

After editing, re-run pnpm sync-language-files and confirm the touched languages have zero remaining TODO translate markers. Reuse the scan from step 2 — an empty result means done:

pnpm sync-language-files
jq -r '.templates[]' translation-manager/templates.json \
  | sed 's/{LANG}/<lang>/' \
  | xargs grep -c "TODO translate" 2>/dev/null | grep -v ':0$' || echo "clean"

Run npx biome check --write on every template you touched. The project enforces Biome formatting + import order.
Do not commit. Per project policy, leave the working tree dirty until the user approves.

Notes

Edit the .template.* files only. Never edit the plain generated files — they are regenerated.
en.tsx / osmTagToNameMapping-en.ts etc. are the masters; English strings only ever change there.
Keep translations natural for the target language. The Slovak hint is usually close to the intended phrasing, but Italian/German/Hungarian/French translations should match the rest of that template's style, not mirror Slavic word order. For distant targets with no close sibling in the hint set (e.g. fr, whose nearest is Italian), review more carefully.
Quoting: single quotes, trailing commas — match what Biome / the surrounding file already uses.
Run the skill once per language or batch several together — either is fine; the verification scan covers both.

Arguments passed: $ARGUMENTS — optional space-separated list of languages to focus on (e.g. pl cs). If empty, process every sync target (see "Which languages").

/translate-missing

Translate keys marked /* TODO translate; non-english translations: ... */ from the generated locale files into the matching *.template.* files.

Background

Which languages

jq -r '.langs[]' translation-manager/templates.json

If $ARGUMENTS names specific languages, process only those; otherwise process every code from the command above.

Steps

Run pnpm sync-language-files to make sure the TODO markers reflect the current state.

For each language, expand its generated files from templates.json and scan them for TODO blocks. Generated *.tsx / *.ts files are gitignored, so the Grep tool won't see them — go through Bash:

# every generated file for <lang> that still has TODO markers, with counts
jq -r '.templates[]' translation-manager/templates.json \
  | sed 's/{LANG}/<lang>/' \
  | xargs grep -c "TODO translate" 2>/dev/null \
  | grep -v ':0$'

Blocks look like:

/* TODO translate; non-english translations:
 * sk: 'Predvolený'
 * cs: 'Výchozí'
 */
default: 'Route default',

The block carries everything needed: the English value (the last line) and any sibling translations. Translate into <lang> using both as context — don't blindly copy the Slovak.

For each TODO key, add the translation to the matching *.template.* file at the same path inside the messages tree:

Locate the parent object in the template by searching for sibling keys that already exist (the English source shows which siblings to look for).
Use the shape-spread / object-literal style already in the file. Add the new key alongside existing siblings.
If a parent object exists as {} or is missing entirely, fill it in / create it. Some sections (e.g. gallery.stats, wikimediaCommons) are empty placeholders in some templates.
If a top-level section is missing (e.g. access, emergency, entrance in OSM mappings), add it where sibling sections sit.

After editing, re-run pnpm sync-language-files and confirm the touched languages have zero remaining TODO translate markers. Reuse the scan from step 2 — an empty result means done:

pnpm sync-language-files
jq -r '.templates[]' translation-manager/templates.json \
  | sed 's/{LANG}/<lang>/' \
  | xargs grep -c "TODO translate" 2>/dev/null | grep -v ':0$' || echo "clean"

Run npx biome check --write on every template you touched. The project enforces Biome formatting + import order.

Do not commit. Per project policy, leave the working tree dirty until the user approves.

Notes

Edit the .template.* files only. Never edit the plain generated files — they are regenerated.

en.tsx / osmTagToNameMapping-en.ts etc. are the masters; English strings only ever change there.

Keep translations natural for the target language. The Slovak hint is usually close to the intended phrasing, but Italian/German/Hungarian/French translations should match the rest of that template's style, not mirror Slavic word order. For distant targets with no close sibling in the hint set (e.g. fr, whose nearest is Italian), review more carefully.

Quoting: single quotes, trailing commas — match what Biome / the surrounding file already uses.

Run the skill once per language or batch several together — either is fine; the verification scan covers both.

Arguments passed: $ARGUMENTS — optional space-separated list of languages to focus on (e.g. pl cs). If empty, process every sync target (see "Which languages").