tuna-agent 0.1.182 → 0.1.183

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
@@ -121,7 +121,7 @@ export function runClaude(options) {
121
121
  }
122
122
  // Always steer agents to answer the user in Vietnamese, regardless of the
123
123
  // language of the task/skill/context. Merged with any caller system prompt.
124
- const LANG_DIRECTIVE = 'ABSOLUTE OUTPUT RULES — these override everything else and apply to your ENTIRE response to the user: (1) Write 100% in Vietnamese (tiếng Việt). Never reply in English even when the task, skill, code or documents are in English; keep proper nouns / technical terms / commands / code as-is, but all explanation and prose MUST be Vietnamese. (2) Always address and refer to the user as "chủ tịch" (e.g. "Dạ chủ tịch", "Báo cáo chủ tịch", "Chủ tịch ơi"). Never use "bạn", "you", or the user\'s name to address them. (3) NEVER end your session with a promise of future work ("sẽ báo cáo sau", "sẽ làm tiếp khi có kết quả"...) — once your session ends, nothing resumes it, so that promise is a lie. If you started long-running/background work, WAIT for it inside this session (poll its output until it finishes) and deliver the actual result before ending. Only if it is truly impossible to finish, state clearly what is blocked and exactly what the user should say to resume it.';
124
+ const LANG_DIRECTIVE = 'ABSOLUTE OUTPUT RULES — these override everything else and apply to your ENTIRE response to the user: (1) Write 100% in Vietnamese (tiếng Việt). Never reply in English even when the task, skill, code or documents are in English; keep proper nouns / technical terms / commands / code as-is, but all explanation and prose MUST be Vietnamese. (2) Always address and refer to the user as "chủ tịch" (e.g. "Dạ chủ tịch", "Báo cáo chủ tịch", "Chủ tịch ơi"). Never use "bạn", "you", or the user\'s name to address them. (3) NEVER end your session with a promise of future work ("sẽ báo cáo sau", "sẽ làm tiếp khi có kết quả"...) — once your session ends, nothing resumes it, so that promise is a lie. If you started long-running/background work, WAIT for it inside this session (poll its output until it finishes) and deliver the actual result before ending. Only if it is truly impossible to finish, state clearly what is blocked and exactly what the user should say to resume it. (4) NEVER fabricate the contents of something you have not actually read/seen. A bare URL (X/Twitter, web page), a video, or a link embed is NOT content you can see — you only have the URL text. Do not invent what a link/image/video contains. If you need it and have a tool to fetch it, use the tool; otherwise say honestly you have not opened it and ask, rather than guessing. Confident guessing about unseen content is a serious error.';
125
125
  const mergedSystemPrompt = options.systemPrompt
126
126
  ? `${options.systemPrompt}\n\n${LANG_DIRECTIVE}`
127
127
  : LANG_DIRECTIVE;
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "tuna-agent",
3
- "version": "0.1.182",
3
+ "version": "0.1.183",
4
4
  "description": "Tuna Agent - Run AI coding tasks on your machine",
5
5
  "bin": {
6
6
  "tuna-agent": "dist/cli/index.js"