scheduled-automation

Scheduled Automation

Use this skill when the user wants the phone to do something later, every day, on specific weekdays, on workdays, or at a fixed minute interval.

Extraction Workflow

For app-centric natural-language requests, use this two-stage workflow:

1. LLM extraction:
   • extract app_name
   • extract operation
   • extract schedule_phrase
2. Rule parsing:
   • let schedule_app_task parse schedule_phrase
   • convert it into repeat, daily_time, days_of_week, or interval_minutes
3. Ambiguity check:
   • if app_name, operation, or schedule_phrase is still ambiguous
   • let the tool return clarification questions first instead of creating the wrong task

Important boundary:

• This skill is only for creating, listing, or cancelling future schedules.
• When an existing schedule fires, do not call schedule_app_task again.
• The saved execution instruction must be immediate and one-shot, e.g. 打开小红书，然后搜索新闻总结后发给我.
• Never keep words such as 每天晚上8点, 每周三, 定时, or 到点 inside operation; those belong only in schedule_phrase.

Example:

• user request: 每周三早上10点打开小红书搜新闻并总结
• LLM extraction:
  • app_name = 小红书
  • operation = 搜新闻并总结
  • schedule_phrase = 每周三早上10点
• rule parsing result:
  • repeat = weekly
  • days_of_week = [3]
  • daily_time = 10:00

Typical requests:

• "每天晚上12点打开微信给张三发消息"
• "每天中午12点打开小红书搜索 AI 新闻并总结"
• "每周三早上10点打开小红书搜索 AI 新闻并总结"
• "每个工作日上午10点打开企业微信提醒我打卡"
• "每隔45分钟打开某个 App 检查一次状态"
• "明天早上8点自动打开企业微信提醒我打卡"
• "帮我定时打开某个 APP 去做某事"
• "每天晚上扫描相册并更新用户画像"

Preferred Tool

schedule_app_task

Use schedule_app_task as the high-level first choice for natural-language scheduling requests about apps and follow-up operations.

It is better than the lower-level schedule_task when the request is naturally phrased as:

• at what time
• open which app
• then do what

Recommended mapping

If the user says:

• "每天晚上12点打开微信去做 X 操作"

Prefer extracting fields first, then call: