cf-translate

Content Translation — Multilingual Publishing

Translate publication-ready ContentForge content into 15 languages while preserving brand voice integrity, citation accuracy, and SEO optimization. Three localization levels let you control the depth of cultural adaptation.

When to Use

Use /contentforge:cf-translate when you need:

• Translated content for international markets (15 languages)
• Brand-consistent multilingual content that matches your source voice in the target language
• SEO-localized content with keywords adapted for target market search behavior
• Citation-safe translations where URLs, source references, and inline citations remain intact
• Cultural adaptation beyond word-for-word translation (adapted or transcreated levels)

Prerequisite: Content must be produced (or imported) through the ContentForge pipeline first. Raw untreated content should go through /contentforge:create-content before translation.

Supported Languages

Localization Levels

Literal (Level 1)

Word-for-word structural translation. Preserves the exact document structure, sentence order, and paragraph boundaries. Translates meaning accurately but does not adapt cultural references, humor, or idiomatic expressions.

Best for: Technical documentation, legal disclaimers, regulatory content, data-heavy research papers.

Adapted (Level 2) — Recommended

Cultural adaptation with structural fidelity. Adjusts cultural references (dates, currencies, idioms, humor) to resonate in the target market while preserving the original content structure and argument flow.

Best for: Articles, blog posts, whitepapers, marketing content for established global brands.

Transcreated (Level 3)

Reimagined for the target market. Preserves the core message and intent but rebuilds the content to feel native in the target language. May restructure sections, replace examples, and adjust tone for maximum local impact.

Best for: Advertising copy, brand storytelling, thought leadership targeting specific regional audiences.

Required Inputs

Minimum Required:

• Source Content -- Google Drive URL, file ID, or local .docx path (must be ContentForge output)
• Target Language -- Language code from the supported list (e.g., es, fr, de)
• Brand -- Brand profile name (must exist, source language profile required)

Optional:

• Localization Level -- literal, adapted, or transcreated (defaults to adapted)
• Regional Variant -- Specify when multiple variants exist (e.g., es-latam, pt-br, fr-ca)
• Target Keywords -- Override SEO keywords for target market (auto-researched if not provided)
• Glossary Override -- Brand-specific term translations that override defaults

How to Use

Interactive Mode

/contentforge:cf-translate

Prompts you for:

1. Source content (URL or file path)
2. Target language (select from 15 options)
3. Brand profile
4. Localization level (literal / adapted / transcreated)
5. Regional variant (if applicable)

Quick Mode

/contentforge:cf-translate "https://drive.google.com/file/d/ABC123" --lang=es --brand=AcmeMed --level=adapted

Multi-Language Batch

/contentforge:cf-translate "https://drive.google.com/file/d/ABC123" --lang=es,fr,de,pt --brand=AcmeMed --level=adapted

Queues translations for all specified languages and processes them sequentially.

What Happens

Step 1: Source Analysis

• Loads source content and brand profile
• Identifies content type, word count, citation count, SEO keywords
• Detects source language (auto-detected or confirmed)
• Quality Gate: Source must be ContentForge output with quality score >= 7.0

Step 2: Element Classification

• Separates content into translatable and immutable elements
• Translatable: Body text, headings, meta tags, alt text, CTAs
• Immutable: Citation URLs, DOIs, proper nouns (configurable), brand names, code snippets, email addresses, phone numbers
• Creates element map with translation instructions per element
• Quality Gate: All elements classified, immutable list confirmed

Step 3: Translation Execution

Dispatch the agent — do not translate inline. Call Task with subagent_type: contentforge:translator, the same way the main orchestrator dispatches its pipeline phases. The Task prompt carries only the source-content path, the brand-profile path, the element map from Step 2, the target language(s) and the localization level. The agent Reads the content itself — never inline a full article into the Task prompt. agents/11-translator.md owns the translation rules, brand-voice mapping, citation preservation and the per-language quality gate; this skill owns argument parsing, source validation and result presentation.

• Translator Agent (11-translator) processes content section by section
• Applies localization level rules (literal / adapted / transcreated)
• Uses DeepL MCP if available (optional) -- falls back to built-in translation
• Preserves document structure, heading hierarchy, and formatting
• Quality Gate: All translatable elements processed, document structure intact

Step 4: Brand Voice Mapping

• Loads brand voice mapping from config/multilingual-patterns.json
• Maps source voice characteristics to target language equivalents
  • Example: "authoritative" in English maps to "formal, datos primero" in Spanish
• Applies target language formality defaults
• Adjusts personality markers for cultural appropriateness
• Quality Gate: Brand voice rating >= 8/10 in target language

Step 5: Citation Preservation Check

• Verifies all citation URLs remain unchanged
• Translates article and book titles in bibliography (preserving original in brackets)
• Confirms inline citation formatting matches source pattern
• Verifies DOIs, ISBNs, and reference identifiers are untouched
• Quality Gate: Zero citation URL changes, zero broken references

Step 6: SEO Adaptation

• Researches target market keywords (or uses provided overrides)
• Translates meta title (<= 60 chars in target language)
• Translates meta description (<= 155 chars in target language)
• Monitors keyword density for target language norms (advisory only — never pad copy to hit a number)
• Generates localized URL slug
• Quality Gate: Meta tags within char limits, and the source's keyword placements (title, H1, first 100 words, 2-3 H2s, conclusion) preserved in the target language. Density is reported, not gated.

Step 7: Target Language Humanization

• Runs Humanizer Agent for target language fluency
• Removes AI telltale phrases specific to target language (loaded from config/multilingual-patterns.json)
• Checks sentence variety (burstiness) against target language norms
• Ensures natural reading flow for native speakers
• Quality Gate: ai_signal_score <= 0.3 for the target language (burstiness is measured and reported but advisory — no minimum to hit, matching the Phase 6.5 humanizer)

Step 8: Quality Verification

• Readability check calibrated for target language (grade level equivalents)
• Brand voice consistency rating (target: >= 8/10)
• Citation integrity confirmation (zero errors)
• SEO keyword placements preserved (density reported as advisory drift, not gated)
• Back-translation spot check (3-5 key sentences translated back to source for meaning verification)
• Quality Gate: All checks pass, composite translation score >= 8.0/10

Step 9: Output

• Generates translated .docx with proper formatting
• Delivers via the brand's tracking backend (tracking.backend in the brand profile): Google Drive folder, Airtable attachment, or local filesystem (~/Documents/ContentForge/{brand}/{type}/{YYYY-MM}/{slug}_{lang}.docx)
• Creates translation report with quality metrics
• Adds a translation entry to the tracking backend

Output

Translated Content Package — SYNTHETIC EXAMPLE, fabricated for illustration:

Translation Complete: "AI in Healthcare: 2026 Trends" --> Spanish (es)

Processing Time: 16 minutes
Translation Score: 8.8/10

Localization Level: Adapted
Regional Variant: es (Spain, neutral)

Quality Metrics:
- Brand Voice Consistency: 9.0/10 (authoritative mapped to "formal, datos primero")
- Citation Integrity: 100% (14 URLs unchanged, 14 titles translated)
- SEO Adaptation: 8.5/10 (meta tags localized, all keyword placements preserved; density 2.0% — advisory)
- Readability: 8.8/10 (Grade 11 equivalent for Spanish)
- Humanization: 9.0/10 (burstiness 0.74, ai_signal_score 0.18)

Content Stats:
- Source Word Count: 1,947
- Translated Word Count: 2,134 (Spanish typically +10% vs English)
- Citations Preserved: 14/14 (100%)
- Immutable Elements: 23 (all preserved)
- Back-Translation Check: 5/5 sentences verified

Output Location:
Google Drive: My Drive/ContentForge Output/AcmeMed/article/2026-07/AI-in-Healthcare-2026-Trends_es_v1.0.docx

Translation Report:
Google Drive: My Drive/ContentForge Output/AcmeMed/article/2026-07/AI-in-Healthcare-2026-Trends_es_translation-report.json

Brand Voice Mapping Examples

The Translator Agent uses config/multilingual-patterns.json to map voice characteristics:

Immutable Element Handling

These elements are never translated:

MCP Integrations

Optional (none required)

• Google Drive -- Only needed if the brand's tracking backend is Google Sheets + Drive, or when running in Cowork (sandbox persistence). Local and Airtable backends work without it.
• DeepL -- Machine translation baseline. If a DeepL MCP server is configured in .mcp.json, the Translator Agent uses its output as a starting point and refines for brand voice, cultural adaptation, and SEO. Before recommending a DeepL MCP package, verify it exists and is maintained on npm (npm view <package> version). When unavailable, the Translator Agent handles all translation natively -- this is the default and works fine.

Relative Effort

Literal is the fastest level; adapted adds cultural-adaptation passes; transcreated is the most thorough and may restructure content. Actual time varies with content length, model speed, and MCP availability — do not promise specific durations.

Limitations

• Source content must be ContentForge output (or at minimum, well-structured markdown with citations)
• RTL languages (Arabic) require downstream layout adjustments in CMS/publishing tool
• Transcreation may change content structure significantly -- review recommended
• Regional variants (es-latam vs es-es) affect vocabulary but not pipeline structure
• Free tiers of translation connectors (e.g., DeepL) have character limits -- monitor usage for high-volume batches

Troubleshooting

"Brand profile not found"

Error: Brand profile "{brand}" not found at
  ~/.claude-marketing/{brand-slug}/Brand-Guidelines/{BrandName}-brand-profile.json
Action: Translation requires a brand profile for voice mapping.
  Run /contentforge:brand-setup "{brand}" first, then retry.

"Source quality score below threshold"

Cause: Source content scored < 7.0 in the original pipeline. Solution: Re-run /contentforge:create-content or /contentforge:content-refresh on the source first.

"Brand voice mapping not found for [language]"

Cause: Target language not configured in config/multilingual-patterns.json. Solution: Check supported languages list. If the language is supported but mapping is missing, update the config.

"Citation count mismatch after translation"

Cause: A citation was accidentally merged or split during translation. Solution: Translator Agent will auto-retry the affected section. If persistent, check for unusual citation formats in the source.

"Meta title exceeds character limit in target language"

Cause: Target language is more verbose than English (common for German, Spanish). Solution: Translator Agent auto-shortens while preserving primary keyword. Review the shortened version.

Related Skills

• /contentforge:create-content -- Produce source content (prerequisite)
• /contentforge:batch-process -- Batch translate multiple pieces
• /contentforge:content-refresh -- Update source before translating
• /contentforge:cf-social-adapt -- Adapt translated content for social platforms

Agents: Translator (11-translator), Humanizer (06.5-humanizer); pseudocode reference: utilities/translation-manager.md (prose doc, not a script) Quality Guarantee: Brand voice >= 8/10, zero citation errors, SEO keywords adapted for target market