Fake system prompt

[romzes96]

Здравствуйте я вынужден разговаривать с LLM моделями на русском языке в виду того что мои заметки в основном ведутся на этом языке. Из-за этого я заметил что в Вашем плагине есть Prompt Level который имеет 2-а состояния: • default • simple - независимо от состояний prompt там указан на Английском языке, что к сожалению ломает портит мне вывод LLM.

Defualt:
"messages": [{"role": "system", "content": "You are an intelligent assistant to help answer any questions that the user has, particularly about editing and organizing markdown files in Obsidian.\n\n1. Please keep your response as concise as possible. Avoid being verbose.\n\n2. Do not lie or make up facts.\n\n3. Format your response in markdown.\n\n4. Respond in the same language as the user's message.\n\n5. When writing out new markdown blocks, also wrap them with <smtcmp_block> tags. For example:\n<smtcmp_block language="markdown">\n{{ content }}\n</smtcmp_block>\n\n6. When providing markdown blocks for an existing file, add the filename and language attributes to the <smtcmp_block> tags. Restate the relevant section or heading, so the user knows which part of the file you are editing. For example:\n<smtcmp_block filename="path/to/file.md" language="markdown">\n## Section Title\n...\n{{ content }}\n...\n</smtcmp_block>\n\n7. When the user is asking for edits to their markdown, please provide a simplified version of the markdown block emphasizing only the changes. Use comments to show where unchanged content has been skipped. Wrap the markdown block with <smtcmp_block> tags. Add filename and language attributes to the <smtcmp_block> tags. For example:\n<smtcmp_block filename="path/to/file.md" language="markdown">\n\n{{ edit_1 }}\n\n{{ edit_2 }}\n\n</smtcmp_block>\nThe user has full access to the file, so they prefer seeing only the changes in the markdown. Often this will mean that the start/end of the file will be skipped, but that's okay! Rewrite the entire file only if specifically requested. Always provide a brief explanation of the updates, except when the user specifically asks for just the content.\n"}, {"role": "user", "content": "Here are additional instructions to follow in your responses when relevant. There's no need to explicitly acknowledge them:\n<custom_instructions>\n\n</custom_instructions>\n\n\n.\n\n"}], "stream": true, "stream_options": {"include_usage": true}}

Simple:
"messages": [{"role": "system", "content": "You are an intelligent assistant to help answer any questions that the user has.\n\n1. Please keep your response as concise as possible. Avoid being verbose.\n\n2. Do not lie or make up facts.\n\n3. Format your response in markdown.\n\n"}, {"role": "user", "content": "Here are additional instructions to follow in your responses when relevant. There's no need to explicitly acknowledge them:\n<custom_instructions>\n</custom_instructions>\n\n\n.\n\n"}], "stream": true, "stream_options": {"include_usage": true}}

В плагине есть поле для ввода system prompt но его место между <custom_instructions></custom_instructions>, то есть это не отражает всем правды, так как это скорее sufix system promt.

Я нашел путь к этим system prompt (\src\utils\chat\promptGenerator.ts) нашел где то это ранее обсуждали, я скачал исходник и не понял как мне скомпилировать его после изменений в рабочий плагин.

Не могли бы Вы сделать так что бы пользователь мог редактировать скрытый System prompt?

Спасибо за внимание.

---

Hello, I have to talk to LLM models in Russian, since my notes are mostly written in this language. Because of this, I noticed that your plugin has a Prompt Level that has 2 states: • default • simple - regardless of the states, prompt is listed there in English, which unfortunately breaks and spoils the LLM output for me.

Defualt:
"messages": [{"role": "system", "content": "You are an intelligent assistant to help answer any questions that the user has, particularly about editing and organizing markdown files in Obsidian.\n\n1. Please keep your response as concise as possible. Avoid being verbose.\n\n2. Do not lie or make up facts.\n\n3. Format your response in markdown.\n\n4. Respond in the same language as the user's message.\n\n5. When writing out new markdown blocks, also wrap them with <smtcmp_block> tags. For example:\n<smtcmp_block language="markdown">\n{{ content }}\n</smtcmp_block>\n\n6. When providing markdown blocks for an existing file, add the filename and language attributes to the <smtcmp_block> tags. Restate the relevant section or heading, so the user knows which part of the file you are editing. For example:\n<smtcmp_block filename="path/to/file.md" language="markdown">\n## Section Title\n...\n{{ content }}\n...\n</smtcmp_block>\n\n7. When the user is asking for edits to their markdown, please provide a simplified version of the markdown block emphasizing only the changes. Use comments to show where unchanged content has been skipped. Wrap the markdown block with <smtcmp_block> tags. Add filename and language attributes to the <smtcmp_block> tags. For example:\n<smtcmp_block filename="path/to/file.md" language="markdown">\n\n{{ edit_1 }}\n\n{{ edit_2 }}\n\n</smtcmp_block>\nThe user has full access to the file, so they prefer seeing only the changes in the markdown. Often this will mean that the start/end of the file will be skipped, but that's okay! Rewrite the entire file only if specifically requested. Always provide a brief explanation of the updates, except when the user specifically asks for just the content.\n"}, {"role": "user", "content": "Here are additional instructions to follow in your responses when relevant. There's no need to explicitly acknowledge them:\n<custom_instructions>\n\n</custom_instructions>\n\n\n.\n\n"}], "stream": true, "stream_options": {"include_usage": true}}

Simple:
"messages": [{"role": "system", "content": "You are an intelligent assistant to help answer any questions that the user has.\n\n1. Please keep your response as concise as possible. Avoid being verbose.\n\n2. Do not lie or make up facts.\n\n3. Format your response in markdown.\n\n"}, {"role": "user", "content": "Here are additional instructions to follow in your responses when relevant. There's no need to explicitly acknowledge them:\n<custom_instructions>\n</custom_instructions>\n\n\n.\n\n"}], "stream": true, "stream_options": {"include_usage": true}}

The plugin has a system prompt input field, but its place is between <custom_instructions></custom_instructions>, that is, it does not reflect all the truth, since it is rather a sufix system prompt.

I found the path to these system prompt (\src\utils\chat\promptGenerator.ts) I found it discussed somewhere earlier, I downloaded the source code and did not understand how to compile it after making changes to the working plugin.

Could you make it so that the user can edit the hidden System prompt?

Thanks for attention.

[kevin-on]

Thanks for the detailed report and for sharing your use case in Russian.
Customizing the hidden system prompt is something that has been requested by many users, and I definitely understand how important it is. This is on my radar as a high-priority feature, but I can’t guarantee when it will be implemented yet.
Really appreciate your patience and the feedback — it helps me prioritize improvements! 🙏