Tencent Hy-MT2 Translator

Problem

Translating technical or product content is rarely a plain text swap: terminology must stay consistent, style may need to be formal, academic, or colloquial, structured data such as JSON, HTML, and Markdown must preserve its format, delimiters may need to remain intact, and context may be required. Calling translation APIs manually means repeatedly handling backend selection, credentials, mode flags, and batch input.

How It Works

The skill wraps the HuanYuan translation model into a workflow:
- Backend selection: supports tencent_cloud or private_model; the private backend must expose an OpenAI-compatible /v1/chat/completions endpoint.
- Credential handling: reads or prompts for the active backend’s model name and API key without re-asking for stored values.
- Mode routing: supports basic translation, terminology constraints, style control, delimiter preservation, structured-data translation, and context-aware translation.
- Batch processing: reads JSONL input, translates line by line, adds a translation field, and supports resumable output.
- Result handling: returns single translations directly, reports batch success/failure counts and output path, and does not expose raw API keys.

Boundaries and Notes

The target language must be among the 38 supported languages. Structured translation depends on clean input; terminology, style, and context are prompt-level constraints and do not guarantee full domain understanding. A private backend must provide a compatible endpoint, model name, and API key.