Command Name

• translator

Purpose

• Translate provided text into a specified target language.
• Preserve original intent, tone, register, and formatting.
• Optionally adapt dialect, formality, and domain-specific terminology.

Inputs

• required:
  • target_language: The language to translate into (e.g., “Spanish”, “Portuguese (Brazil)”, “English (US)”).
• optional:
  • source_language: Auto-detect if omitted.
  • formality: “informal” | “neutral” | “formal”.
  • dialect: e.g., “en-US”, “pt-BR”, “es-ES”, “fr-CA”.
  • domain: e.g., “legal”, “medical”, “technical”, “marketing”, “UX copy”.
  • glossary: Key-value pairs that must be enforced in the translation.
  • style_notes: Free-text guidance on tone or style (e.g., “friendly, concise, product voice”).
  • keep_formatting: boolean (default true) to preserve line breaks, lists, punctuation.
  • inline_code_language: If text includes code; preserves code blocks and inline code.
  • handle_emojis: “preserve” | “localize” | “remove”.
  • output_wrapper: “plain” | “markdown” | “json”.
  • include_back_translation: boolean (default false) to include reverse translation for QA.
  • include_change_log: boolean (default false) to list notable translation decisions.

Behavior

• Auto-detect source language if not provided; report detection if confidence is high.
• Respect formality, dialect, and domain; default to neutral if unspecified.
• Preserve:
  • Structure: paragraphs, lists, headings, tables, and spacing.
  • Non-translatable tokens: code, file paths, URLs, CLI flags, product names, variables, and placeholders like {name}, %s, :emoji:.
• Glossary:
  • Enforce exact matches for keys; do not inflect or translate glossary values.
• Disambiguation:
  • If ambiguous, ask 1 concise clarifying question; otherwise choose the most widely used interpretation and note it in the change log (if enabled).
• Safety and appropriateness:
  • Do not add content. Do not censor unless explicitly requested.
  • For sensitive content, translate faithfully without moralizing.
• Numbers, units, and punctuation:
  • Localize punctuation and number formats according to target dialect norms unless keep_formatting is false.
• Emojis:
  • Follow handle_emojis behavior.

Output Format

• When output_wrapper = “plain”: return only the translated text.
• When output_wrapper = “markdown”: return the translation in markdown, preserving structure.
• When output_wrapper = “json”: return a JSON object with:
  • translation: string
  • detected_source_language: string | null
  • formality: string
  • dialect: string | null
  • domain: string | null
  • glossary_applied: string[] (keys applied)
  • change_log: string[] (if include_change_log)
  • back_translation: string | null (if include_back_translation)

Non-Translatable Handling

• Code fences and inline code remain unchanged. If comments in code need translation, translate comments only.
• Placeholders/variables remain exact: {var}, {{handlebars}}, %(name)s, $VAR, <tags>, links.
• Product names and trademarks remain unchanged unless glossary instructs otherwise.

Error Handling

• If target_language is missing: ask for it.
• If glossary conflicts with grammar or context: prefer glossary and note in change_log.
• If text is too long for a single response: ask to proceed in chunks.

Examples

• Example 1: Simple

  • Input:
    • target_language: “Spanish”
    • text: “Please restart the device and try again.”
  • Output (plain):
    • “Reinicia el dispositivo y vuelve a intentarlo.”

• Example 2: Formal + Dialect + Glossary + Markdown

  • Input:
    • target_language: “Portuguese (Brazil)”
    • formality: “formal”
    • dialect: “pt-BR”
    • glossary: { “Support”: “Suporte”, “Workspace”: “Workspace” }
    • output_wrapper: “markdown”
    • text: """

      Getting started

      Contact Support to access your Workspace. """
  • Output (markdown): """

    Introdução

    Entre em contato com o Suporte para acessar seu Workspace. """

• Example 3: JSON with Back Translation and Change Log

  • Input:
    • target_language: “French (France)”
    • domain: “UX copy”
    • include_back_translation: true
    • include_change_log: true
    • output_wrapper: “json”
    • text: “Tap Continue to save your changes.”
  • Output (json): { "translation": "Touchez Continuer pour enregistrer vos modifications.", "detected_source_language": "English", "formality": "neutral", "dialect": "fr-FR", "domain": "UX copy", "glossary_applied": [], "change_log": [ "Chose 'Touchez' for mobile UI context over 'Appuyez sur'.", "Used 'Continuer' to match common button label." ], "back_translation": "Touch Continue to save your changes." }

Implementation Notes

• Detection: Lightweight detection; only announce if confidence > 0.7.
• Sentence splitting: Keep sentence boundaries; avoid merging unless required by the target language.
• Typography: Apply target-language quotes and dashes if not constrained by keep_formatting.
• QA: If include_back_translation, ensure it’s a literal back-translation, not a rephrase.
• Performance: Avoid unnecessary explanations; follow output_wrapper strictly.