Prompt Translator Java Edition

Problem: Generic Prompts Fall Short

When users say “help me plan, analyze, or decide,” models often return generic templates. In Chinese requests, intent is frequently hidden in casual wording, omissions, and domain context, so a direct LLM pass can produce answers that are hard to execute. This Java skill turns “prompt translation” into a local workflow: it identifies intent, loads domain rules, retrieves supporting information, and outputs structured results, reducing token usage and network dependence.

How It Works: Intent, Rules, and RAG

• Intent recognition: Determines whether the user wants feasibility advice, multi-dimensional analysis, workflow planning, or a direct answer.
• Dynamic domain loading: Applies rules for the identified domain instead of forcing one generic prompt into every case.
• RAG retrieval: The local version supports automatic retrieval and full caching; the SKILL platform version may require manual triggering and has limited caching.
• Output constraints: Results are generated according to structured norms, making them easier for downstream agents or business systems to consume.
• M1 research mode: When requirements are ambiguous, the system asks for missing fields step by step and writes them into JSON rather than guessing.

Boundaries

It is best suited for Chinese-language prompt rewriting, intent clarification, and structured generation, not for arbitrary code generation or as a complete business system. If you need automatic RAG, caching, and fully open rule extension, the local ZIP version is preferred; the platform version has limitations.