tts-task-notify

TTS Task Completion Notification

You are equipped with a voice notification system. A Stop hook monitors your output and automatically plays a voice summary via a TTS server on the local network. You must intelligently decide when to trigger it.

When to Notify — Smart Decision Tree

Use the following decision tree to determine whether to output a voice summary. Evaluate from top to bottom, first match wins:

1. Always Notify (high confidence)

• Completed a task that modified files or system state (code edits, install, deploy, config changes)
• Completed a multi-step debugging, research, or analysis session
• User has been waiting — the task took significant time or multiple tool calls
• User gave a direct instruction ("帮我改...", "部署一下", "修复这个bug") and you finished it

2. Notify with Voice Reply (conversational)

Even without "real work", notify when your reply is a substantive answer the user would benefit from hearing:

• User asked a specific question and you gave a clear, actionable answer (e.g., "这个错误是因为..." "你应该用...")
• User asked you to check, verify, or confirm something and you have a definitive result (e.g., "服务正常" "端口没开")
• Your reply contains a conclusion or recommendation the user needs to act on
• The conversation feels like voice chat — short, interactive, back-and-forth exchanges

For these cases, summarize your answer in spoken form rather than summarizing "work done".

3. Do NOT Notify (skip)

• Clarifying questions — you're asking the user for more info, not answering
• Partial progress — intermediate steps of a larger task (wait until the end)
• Echoing back — merely confirming what the user said without adding value ("好的，明白了")
• Showing raw data — dumping logs, listing files, or displaying status without interpretation
• Meta-conversation — discussing how to approach a task rather than doing it

Decision Heuristic

Ask yourself: "If the user were across the room, would they want to hear this?"

• Yes, they'd want to know -> Notify
• They'd need to see the screen anyway -> Skip

Summary Format

At the end of your response (after all code, explanations, etc.), output exactly:

【任务完成总结】简短的完成摘要[[END_SUMMARY]]

The markers MUST be output as raw text, NOT inside a fenced code block.

Rules

1. Content: 100 characters max, natural spoken Chinese. For tasks: describe what was done and the result. For voice replies: summarize the key answer or conclusion
2. No code: Never include code, commands, file paths, or technical jargon in the summary
3. Spoken style: Write as if speaking aloud — "已完成" not "done", "三个接口" not "3 APIs"
4. Frequency: At most ONE summary per response, only at the very end
5. Match language: Use the same language the user is speaking (typically Chinese)

Examples — Notify

Task completion (Category 1):

【任务完成总结】用户认证模块已完成，包括登录、注册和密码重置三个接口，所有测试通过。[[END_SUMMARY]]

【任务完成总结】数据库迁移脚本已执行成功，新增了订单表和支付记录表。[[END_SUMMARY]]

【任务完成总结】前端首页重构完成，加载速度从三秒优化到一点二秒。[[END_SUMMARY]]

Voice reply (Category 2) — summarize the answer, not "work done":

【任务完成总结】这个报错是因为数据库连接超时，建议把超时时间从五秒改成三十秒。[[END_SUMMARY]]

【任务完成总结】服务器状态正常，三个后端都在运行，内存占用百分之四十五。[[END_SUMMARY]]

【任务完成总结】推荐用FastAPI，性能好，自带文档生成，比Flask更适合你的场景。[[END_SUMMARY]]

Examples — Do NOT Notify

• You ask: "你希望用哪个数据库？" -> Skip (clarifying question)
• User: "继续" -> You do step 2 of 5 -> Skip (partial progress)
• You say: "好的，我来处理" -> Skip (echoing, no substance)
• You dump 50 lines of log output -> Skip (raw data, need screen)

TTS Health Check

You can verify the TTS server is reachable:

curl -s "${TTS_SERVER_URL:-http://localhost:58201}/health" | head -c 200

Available Voices

edge-tts (recommended for speed): xiaoxiao, xiaoyi, yunjian, yunxi, yunxia, yunyang, Serena (auto-mapped)

qwen3-tts (highest quality): Serena, Vivian, Dylan, Eric, Ryan, Aiden, Uncle_Fu, Ono_Anna, Sohee

cosyvoice3 (voice cloning): 中文女, 中文男, 英文女, 英文男, 日语男, 粤语女, 韩语女