german-translator-pro-brand-legal

Swiss-German translator for brand/UI and legal text — trained on real translation patterns from 12+ top brands (Swisscom, UBS, On Running, Logitech, IKEA, Stripe, Notion, Slack, Apple, Shopify, BMW, Vercel). Translates and validates i18n JSON or plain text from English to Swiss-formal German (Hochdeutsch CH). Anti-translationese: restructures German thoughts fresh instead of mapping English. Two text types: brand/UI and legal. Two actions: TRANSLATE and VALIDATE. Includes 100+ brand-confirmed translation pairs, 40+ glossary terms with keep/translate decisions, DSGVO/DSG legal terminology, completeness checks, and a 4-agent pipeline for legal documents.

Safety Notice

This listing is imported from skills.sh public index metadata. Review upstream SKILL.md and repository scripts before running.

Copy this and send it to your AI assistant to learn

Install skill "german-translator-pro-brand-legal" with this command: npx skills add SLFMR1/german-translator-pro-brand-legal

German Translator Pro — Brand & Legal

Translate and validate English to Swiss-formal German (Hochdeutsch CH). Trained on real translation patterns crawled from 12+ top brands — not generic machine translation.

What makes this different

LLMs produce translationese — grammatically correct German that sounds translated, not written. This skill fixes that with curated reference files extracted from how top brands actually translate their websites, marketing, and legal pages.

Brands crawled: Swisscom, UBS, On Running, Logitech, IKEA CH, Stripe, Notion, Slack, Apple, Shopify, BMW, Vercel

What we extracted: 100+ real EN→DE pairs, 5 hero headline strategies, 15+ CTA patterns, Du/Sie industry split, 40+ glossary decisions (keep vs translate), 29 legal term translations, translationese traps, legal register patterns — all with brand evidence.

The agent reads these references before every translation, producing output that matches how these brands write German.

Install

npx skills add SLFMR1/german-translator-pro-brand-legal

When to Use

  • Translating UI strings (buttons, headings, CTAs, marketing copy) to German
  • Translating legal pages (privacy policy, terms of service, imprint) to German
  • QA-checking existing German translations against English source
  • Any EN → DE translation that needs Swiss conventions and natural phrasing

How It Works

The skill supports two text types and two actions:

Brand / UILegal
Translate (EN → DE)Natural, punchy Swiss-German that sounds written, not translatedPrecise, formal German with DSGVO/DSG legal terminology
Validate (QA check)Check for naturalness, accuracy, Swiss conventionsCheck for legal precision, qualifier preservation, terminology

Key features

  • Anti-translationese — Restructures German thoughts fresh instead of mapping English sentence structures. Trained on how top brands actually write
  • Swiss (CH) or German (DE) locale — Swiss: ss not ß, 1'000, CHF. German: standard ß, 1.000, EUR. User chooses
  • Back-translation QA — Verifies meaning preservation by translating DE back to EN
  • Completeness checks — Catches untranslated strings AND accidentally germanized brand terms
  • Legal term precision — 25+ DSGVO/DSG terms with brand-confirmed translations and source fidelity rules
  • 4-agent legal pipeline — Translator → Back-Translator → Validator → Refiner for legal documents
  • Configurable glossary — Keep-in-English terms, always-translate terms, split decisions — all with brand evidence

Workflow

Step 1: Determine Text Type, Action, Address Form, and Locale

Before doing any work, determine four things. If not clear from context, ask the user:

1. Text type?

  • UI — hero sections, CTAs, agent descriptions, navigation, headings, buttons, labels
  • LEGAL — privacy policy, terms of service, imprint, legal disclaimers

2. Action?

  • TRANSLATE — convert English to German
  • VALIDATE — QA-check existing German translation against English source

3. Address form?

  • Sie (formal) — B2B, corporate, legal, professional. Capitalized: Sie, Ihnen, Ihr, Ihre
  • Du (informal) — B2C apps, casual products, startups. Lowercase: du, dir, dein, deine

Default to Sie if unclear. Legal text always uses Sie regardless of user preference.

4. Locale?

  • Swiss (CH)ss not ß, apostrophe thousands (1'000), CHF before amount (CHF 100), decimal point (99.90)
  • German (DE)ß where standard, dot thousands (1.000), € after amount (100 €), decimal comma (99,90)

Default to Swiss (CH) if unclear.

Apply the chosen form and locale consistently throughout. Follow the matching section below.

Web research allowed. If a term is not in the glossary, or you're unsure about the right translation, look it up. Fetch the German version of a brand's website (e.g. apple.com/de, stripe.com/de, notion.so/de) to see how they actually translate it. You can also check other relevant brand websites in the same industry as the user's project. Real-world usage beats guessing.

Step 2: Multi-File Projects — Use Subagents

When translating or validating multiple files or pages, do NOT process them all in the main context. Instead, use the orchestrator/subagent pattern:

Orchestrator (main agent):

  1. Determine the 4 settings above (text type, action, address form, locale)
  2. List all files/pages to process
  3. Spawn one subagent per file/page using the Agent tool
  4. Collect results and report a summary

Each subagent:

  1. Receives: the file path, text type, action, address form, locale
  2. Reads the reference files (glossary, style guide, examples)
  3. Translates or validates the single file
  4. Self-validates against qa-checklist.md — runs the relevant checks for the text type
  5. Fixes any CRITICAL issues found during self-validation
  6. Reports back: the output file + a summary (PASS/FAIL, issue count, any unresolved warnings)

Why: Each subagent gets a clean context with full reference files loaded. No context blowup from processing 10+ files sequentially. Self-validation catches issues before they reach the orchestrator.

Single file? Skip this — just run the translation/validation directly.


TRANSLATE + UI

Natural, punchy Swiss-German for marketing and interface text.

Process

  1. Read the English source JSON file completely
  2. Read glossary.md for terms that must stay untranslated
  3. Read swiss-german-style.md for tone, conventions, and translationese traps
  4. Read examples.md — match this quality level
  5. Translate every value. Never modify keys or {variable} placeholders
  6. Self-check: Read each German value out loud. Would a Swiss product manager say this over coffee? If it sounds like a document, rewrite it
  7. Back-translate check: For 5 random keys, mentally translate the German back to English. If the meaning shifted, fix it
  8. Completeness check: Compare EN keys vs DE keys — zero missing. Scan for identical EN/DE values: keep-in-English terms are OK, but full English sentences left untranslated are not. Verify always-translate terms from glossary are actually translated (Products→Produkte, not "Products")
  9. Length check: Compare character count of each DE value vs EN value. Flag any UI string (buttons, labels, nav, headings, tooltips) where DE is >120% of EN length. Shorten these — German must fit the same visual space
  10. Output a complete JSON file with identical structure to the English source

Rules

  • Restructure, don't translate. Write the German thought fresh. If it reads like it was run through a translator, rewrite it
  • Use the address form chosen in Step 1 (Sie or Du) — apply consistently throughout
  • Apply the locale conventions chosen in Step 1 (CH or DE) for ß/ss, number formatting, and currency
  • Preserve all {variable} placeholders and HTML tags exactly
  • Keep JSON key names and namespace structure identical
  • Brand names, product names stay in English (see glossary.md)
  • English tech terms standard in Swiss business stay English (Agent, CRM, API, Inbox, Tools, Workflow, etc.)
  • Prefer verbs over nominalizations — "prüfen" not "Durchführung der Prüfung"
  • Prefer concrete benefits over abstract claims
  • Short sentences. If over 20 words, consider splitting
  • Length constraint (UI critical). German text expands ~20-30% vs English. UI strings must fit the same visual space — buttons, labels, nav items, headings, tooltips. If the German is noticeably longer than the English, shorten it. Prefer compact phrasing: "Jetzt starten" not "Jetzt loslegen und starten", "Mehr erfahren" not "Erfahren Sie mehr darüber". For buttons and nav: aim to match or stay within ~120% of the English character count

Output

Complete valid JSON. Same keys, same nesting, German values.


TRANSLATE + LEGAL

Precise, formal Swiss-German for legal documents. Accuracy over style.

For the full multi-agent pipeline (recommended for legal), see legal-workflow.md. It uses 4 agents: Translator → Back-Translator → Validator → Refiner. The instructions below are for the Translator agent (Agent 1).

Process

  1. Read the English source JSON file completely
  2. Read glossary.md for terms that must stay untranslated
  3. Read swiss-german-style.md — apply the locale conventions (CH or DE) chosen in Step 1, but use formal legal register, not marketing tone
  4. Translate every value with focus on legal precision and completeness
  5. Self-check — 4 passes:
    • Pass 1: Back-translate every value to English. If meaning shifted → fix
    • Pass 2: Check every legal qualifier (may/shall/unless) is preserved with correct weight
    • Pass 3: Read only the German text without looking at English. Flag ambiguities
    • Pass 4: Completeness — diff EN keys vs DE keys (zero missing). Scan identical values: brand names and keep-in-English terms OK, untranslated legal text is CRITICAL. Verify legal section headings, consent language, and rights descriptions are all in German
  6. Output two files:
    • The complete translated JSON (same structure as source)
    • A _review_notes.md file flagging sections that need professional legal review

Rules

  • Legal precision over natural flow. Accuracy of meaning is paramount. Do not rephrase for punchiness
  • Formal register. Legal text should sound formal and precise — "stiff" is correct here
  • Formal "Sie" form — always for legal text, regardless of user preference for UI
  • Apply the locale conventions chosen in Step 1 (CH or DE) for ß/ss, number formatting, and currency
  • Preserve all {variable} placeholders and HTML tags exactly
  • Keep JSON key names and namespace structure identical
  • Preserve legal terminology. Use standard DSGVO/DSG terms (see table below)
  • Swiss legal context (CH locale). Reference Swiss law (DSG/nDSG, OR) alongside EU references (DSGVO/GDPR) where the English mentions data protection or privacy. For DE locale, DSGVO references are sufficient
  • Do NOT simplify. Complex sentences keep their complexity. Legal meaning depends on precise wording
  • Do NOT drop qualifiers. "may", "shall", "unless", "notwithstanding" have specific legal weight: "kann", "hat zu" (not "soll" — "soll" is weaker, closer to "should"), "sofern nicht", "ungeachtet"
  • Flag jurisdiction-specific clauses — courts, regulators, applicable law references go in the review notes
  • Source fidelity. Translate what the source says, not what it should say. If the source uses "personal information" (informal), translate as "persönliche Informationen" — do NOT upgrade to "personenbezogene Daten" (which is the DSGVO term for "personal data"). Only use formal legal terms when the source uses the corresponding formal English term
  • No right escalation. If the source says "opt out", translate as "abmelden" or "ablehnen" — do NOT escalate to "widersprechen" (formal DSGVO right of objection under Art. 21). Match the weight of the source
  • Complete compound nouns. German compound truncation ("Rechts- oder Behörden...") must be grammatically complete. "Rechts- oder Behördenanfragen" is correct. "Rechts- oder Behörden" alone is not — the compound base word must appear at least once

Legal Term Translations

EnglishGermanNotes
personal datapersonenbezogene DatenDSGVO standard
data processingDatenverarbeitung
data controllerVerantwortlicher
data processorAuftragsverarbeiter
consentEinwilligungnot "Zustimmung" in GDPR context
legitimate interestberechtigtes Interesse
data subjectbetroffene Person
right to erasureRecht auf Löschung
terms of serviceNutzungsbedingungenor AGB
liabilityHaftung
indemnificationFreistellung / Schadloshaltung
governing lawanwendbares Recht
severabilitySalvatorische Klausel
intellectual propertygeistiges Eigentum
data protectionDatenschutz
data retentionDatenaufbewahrung
data transferDatenübermittlungApple, Stripe
legal basisRechtsgrundlageApple, Stripe
supervisory authorityAufsichtsbehördeApple
Data Protection OfficerDatenschutzbeauftragterApple, Stripe
Standard Contractual ClausesStandardvertragsklauselnApple, Stripe, Slack
right to accessAuskunftsrechtApple
right to rectificationRecht auf BerichtigungApple
right to objectWiderspruchsrechtLogitech
right to data portabilityRecht auf DatenübertragbarkeitLogitech
GDPRDSGVO (Datenschutz-Grundverordnung)Always "DSGVO" in DE, never "GDPR"
FADP / Swiss DPADSG (Bundesgesetz über den Datenschutz)Stripe, IKEA
FDPICEDÖBSwiss federal data protection authority
as is / as availablewie besehen / wie verfügbarShopify — warranty disclaimer

Output

Two files:

  1. Complete valid JSON — same keys, same nesting, German values
  2. _review_notes.md:
# Legal Translation Review Notes: [filename]

## Sections Requiring Professional Legal Review
- [key.path]: Reason

## Translation Decisions
- [key.path]: "Chose X over Y because..."

## Swiss Law Considerations
- Areas where Swiss law (DSG/nDSG, OR) differs from the EU law referenced in source

VALIDATE + UI

QA-check German UI/marketing translations for naturalness, accuracy, and Swiss conventions.

Process

  1. Read both the English source and German translation JSON files
  2. Read glossary.md and swiss-german-style.md
  3. Read examples.md — compare translations against these benchmarks
  4. Read qa-checklist.md — run the Structural, Accuracy, Naturalness, Tone & Style (UI), Linguistic (UI), and Completeness sections
  5. Back-translation check: For each DE value, mentally translate it back to English. Mismatch → CRITICAL (meaning error) or WARNING (nuance drift)
  6. Naturalness check: Does this sound written in German or translated from English? Flag translationese as WARNING
  7. Completeness check: Run the full Completeness routine from qa-checklist.md:
    • Identify all keys where EN value === DE value (identical-value scan)
    • For each, classify: keep-in-English term → OK, brand name → OK, URL/email/number → OK, actual untranslated text → CRITICAL
    • Cross-check against glossary: no accidentally translated keep-in-English terms, no English left for always-translate terms
    • Verify all navigation terms, CTAs, and common UI patterns are translated per brand-crawl evidence
  8. Length check: Compare character count of each DE value vs EN value. Flag UI strings (buttons, labels, nav, headings, tooltips) where DE is >120% of EN length as WARNING — these may break layout
  9. Web crosscheck: For any translation you're unsure about or that feels off, fetch the German version of a relevant brand website to verify how they translate it. Spot-check at least 5 key terms against real brand usage
  10. Report findings grouped by severity

VALIDATE + LEGAL

QA-check German legal translations for precision, terminology, and legal correctness.

If a back-translation file is available (from the legal-workflow.md pipeline), use it in step 4 below. This catches meaning errors that bilingual review misses — a reviewer unconsciously imports the correct meaning from the source, but a back-translation reveals what the German text actually says in isolation.

Process

  1. Read both the English source and German translation JSON files
  2. Read glossary.md and swiss-german-style.md
  3. Read qa-checklist.md — run the Structural, Accuracy, Tone & Style (Legal), Linguistic (Legal), Legal-Specific, and Completeness sections
  4. Back-translation diff (if back-translation file provided): Compare original EN with back-translated EN, key by key. Flag divergences:
    • Meaning matches → OK
    • Slight nuance drift → WARNING
    • Material change (obligation→permission, condition dropped, scope changed) → CRITICAL
  5. Blind read: Read ONLY the German text without looking at English. Flag anything unclear or ambiguous as a standalone German legal document
  6. Precision check: Are legal qualifiers ("may", "shall", "unless") translated with full legal weight? Are complex conditions preserved without simplification?
  7. Completeness check: Run the full Completeness routine from qa-checklist.md:
    • Identical-value scan: classify every key where EN === DE (keep-in-English → OK, brand name → OK, actual untranslated text → CRITICAL)
    • Cross-check against glossary: keep-in-English terms not accidentally translated, always-translate terms not left in English
    • Legal-specific: verify legal section headings, consent language, rights descriptions are all translated — never left in English
  8. Web crosscheck: For any legal term or translation you're unsure about, fetch the German version of a relevant brand's legal pages (e.g. apple.com/de/privacy, stripe.com/de/legal) to verify standard usage. Spot-check at least 5 key legal terms against real brand usage
  9. Report findings grouped by severity

Severity Levels (Both VALIDATE Modes)

  • CRITICAL: Missing keys, broken placeholders, broken JSON, meaning errors, legal inaccuracies
  • WARNING: Tone issues, phrasing problems, terminology inconsistencies, Bundesdeutsch instead of Swiss
  • INFO: Style suggestions, alternative phrasings, minor improvements

Output Format (Both VALIDATE Modes)

## Validation Report: [filename]
### Text type: [UI / LEGAL]

### CRITICAL (X issues)
- [key.path]: Description of issue

### WARNING (X issues)
- [key.path]: Description of issue | Suggestion: ...

### INFO (X suggestions)
- [key.path]: Suggestion

### Summary
X critical | X warnings | X info — [PASS/FAIL]

FAIL if any CRITICAL issues exist. PASS otherwise.


Resources

Source Transparency

This detail page is rendered from real SKILL.md content. Trust labels are metadata-based hints, not a safety guarantee.

Related Skills

Related by shared tags or category signals.

Automation

Experience Iceland Aurora Fever

When the sky breaks open and bleeds light. An immersive journey on drifts.bot — 7 steps, HIGH intensity, 1-2 hours. Browse, start, and travel via API.

Registry SourceRecently Updated
Automation

AgentMade — Directory for Agent-Built Projects

Submit, discover, vote on, and comment on AI agent-built projects on AgentMade — the first public directory dedicated to things built by AI agents. Use this...

Registry SourceRecently Updated
Automation

Video Call AI Character

Video call the user with a real-time AI avatar powered by Runway. The agent initiates calls to the user — for standups, urgent alerts, check-ins, or any conv...

Registry SourceRecently Updated
Automation

Vestibular Tutor

Agente tutor socrático do sistema de preparação para o vestibular de Antonio (FUVEST - Ciência da Computação). Especializado em ensinar qualquer disciplina d...

Registry SourceRecently Updated