diet-pattern-detection

name	diet-pattern-detection
version	1.0.0
description	One-time cron-only skill (never user-triggered): checks if the user's actual eating pattern fits a different diet mode, then self-destructs.
metadata	{"openclaw":{"emoji":"bar_chart","homepage":"https://github.com/NanoRhino/weight-loss-skill"}}

Diet Pattern Detection

⚠️ Cron-only skill. Never triggered by user messages. Activated by a daily cron job created by notification-manager at onboarding completion.

⚠️ Output Format (HARD RULE — Never violate)

When this skill sends a message to the user (diet pattern detection result, mode switch suggestion, or any output), the first line MUST be the section banner based on user locale:

USER.md locale	First line
zh / zh-CN / zh-TW / unset	🔄——饮食模式调整——🔄
en / en-*	🔄——Diet Mode Adjustment——🔄

Then one blank line, then the message body.

✅ Correct (Chinese user):

🔄——饮食模式调整——🔄

📋 我注意到一个有趣的现象...

❌ Wrong — missing banner, or banner not on first line, or no blank line after banner.

Trigger

Daily cron job. Start date: Onboarding Completed + 3 days (from health-profile.md > Automation). Cron time: dinner_time + 3h.

Workflow

1. Check precondition — read health-profile.md > Automation > Pattern Detection Completed. If already has a date → skip (should not happen, but defensive).

2. Run detection (handles data sufficiency check internally):

   python3 {baseDir}/scripts/detect-pattern.py \
     --data-dir {workspaceDir}/data/meals \
     --current-mode <from health-profile.md > Diet Config > Diet Mode> \
     --weight <current weight from latest weight.json entry> \
     --target-weight <target weight from PLAN.md> \
     --tz-offset {tz_offset}
   

   If result has reason: "insufficient_data" (< 3 days) → reply exactly NO_REPLY (suppresses delivery; do NOT emit a free-text explanation). Cron preserved, retry tomorrow.

3. Handle result:

   • has_pattern: true → read references/diet-pattern-response.md, compose and send message to user. If user agrees to switch → update health-profile.md > Diet Config > Diet Mode and recalculate targets.
   • has_pattern: false → no message (pattern matches current mode, all good).

4. Self-destruct (when result is has_pattern true or false, NOT insufficient_data):

   1. List current agent's cron jobs: bash {reminder-manager:baseDir}/scripts/list-reminders.sh --agent {agentId}
   2. Find job with name containing "diet-pattern" or "Diet pattern"
   3. Delete it: bash {reminder-manager:baseDir}/scripts/delete-reminder.sh --agent {agentId} --job <jobId>
   4. Write completion date to health-profile.md > Automation > Pattern Detection Completed: <YYYY-MM-DD>

   ⚠️ Never call the cron tool directly (cron list / cron remove) — always go through the reminder-manager scripts above (AGENTS.md red line; cross-user cron deletion incident).

Reference Files

• references/diet-pattern-response.md — user notification template (when pattern detected)