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:

{
  "action": "create",
  "task_name": "daily-wechat-task",
  "app_name": "微信",
  "operation": "做 X 操作",
  "schedule_phrase": "每天晚上12点"
}

Tool Parameters

schedule_app_task

• action: create | list | cancel
• task_name: task display name
• app_name: target app name, e.g. 微信
• package_name: optional package name for higher precision
• operation: what to do after opening the app
• schedule_phrase: preferred natural-language schedule phrase extracted by the LLM, e.g. 每周三早上10点
• repeat: daily | once | weekly | workday | interval
• time_text: backward-compatible alias of schedule_phrase
• days_of_week: used with repeat=weekly, e.g. ["mon", "wed"] or ["周三"]
• interval_minutes: used with repeat=interval, e.g. 30 or 45
• run_at: one-time target time
• delay_seconds: one-time delay
• task_id: used when cancelling

When to fall back to schedule_task

Use lower-level schedule_task when:

• the task is not app-centric
• the user wants to schedule a generic agent instruction
• the execution target is not just "open app then do something"
• the task is a memory maintenance workflow such as gallery syncing or profile rebuilding

Example: schedule gallery memory sync

If the user says:

• "每天晚上 10 点扫描相册并更新用户画像"

Prefer using schedule_task with a clear instruction that tells the agent to use gallery_memory:

{
  "action": "create",
  "task_name": "daily-gallery-memory-sync",
  "instruction": "使用 gallery_memory 工具同步相册中的新增图片记忆，并更新用户画像。",
  "repeat": "daily",
  "daily_time": "22:00"
}

Best Practices

1. Prefer schedule_app_task for "定时打开 App 并操作" requests.
2. Preserve the user's original business intent inside operation.
3. If the target app is ambiguous, ask for clarification or include package_name.
4. For first-time setup, remind the user to verify the task once on a real device.
5. When the task is critical, suggest checking exact alarm permission and background restrictions.
6. If the extracted fields still look vague, do not force task creation; let the tool ask follow-up questions first.
7. Creation-time query and execution-time instruction are different: creation uses schedule_phrase; execution must contain only the action to perform now.

Example Commands

Daily task

{
  "action": "create",
  "task_name": "daily-xhs-news",
  "app_name": "小红书",
  "operation": "搜索 AI 新闻，并总结前三条结果",
  "schedule_phrase": "每天中午12点"
}

Weekly task

{
  "action": "create",
  "task_name": "weekly-xhs-news",
  "app_name": "小红书",
  "operation": "搜索 AI 新闻，并总结前三条结果",
  "schedule_phrase": "每周三早上10点"
}

Workday task

{
  "action": "create",
  "task_name": "workday-checkin",
  "app_name": "企业微信",
  "operation": "提醒我打卡",
  "schedule_phrase": "每个工作日上午10点"
}

Fixed interval task

{
  "action": "create",
  "task_name": "interval-status-check",
  "app_name": "小红书",
  "operation": "检查一次首页热点并总结",
  "schedule_phrase": "每隔45分钟"
}

One-time task

{
  "action": "create",
  "task_name": "wechat-reminder-tonight",
  "app_name": "微信",
  "operation": "给张三发送消息：明天记得交日报",
  "repeat": "once",
  "delay_seconds": 600
}

Cancel task

{
  "action": "cancel",
  "task_id": "your-task-id"
}

Verification Hint

After creating a task, you can use schedule_app_task(action="list") to confirm:

• task exists
• next trigger time is correct
• generated instruction matches the user's intent