hermes - 💡(How to fix) Fix Support localized system prompt to align model thinking language

Feature Description

When using DeepSeek V4 Pro (or other models), the model's internal reasoning/thinking language defaults to English even when the user communicates entirely in Chinese. This is because Hermes's entire system prompt — tool descriptions, instructions, environment hints, skill content — is in English, anchoring the model to think in English.

On DeepSeek's official web chat, the same model thinks in Chinese because the surrounding context is all Chinese.

Proposed Solution

Provide a display.language setting that translates key system prompt components (tool descriptions, core instructions) into the target language, or at minimum injects a strong language directive into the system prompt to align the model's thinking language with the user's language.

Minimally, a hermes config set display.language zh-CN could:

1. Translate core identity and instruction blocks
2. Add a directive like "Prefer thinking in the same language as the user"
3. Eventually support full tool description localization

Motivation

• Non-English users get a disjointed experience: chat in native language, but model thinks in English
• Thinking in the user's language improves reasoning quality for language-specific tasks
• Other AI products (DeepSeek web, ChatGPT) already align thinking language to UI language

Environment

• WSL2 Ubuntu 26.04
• Hermes Agent (latest)
• DeepSeek V4 Pro via api.deepseek.com