Full Daily Planner

Complete daily life management: health data import → health report → next-day plan → system crontab reminders.

Architecture

15:00 agentTurn (isolated) — Health Report
  ├── runs sleep_import.py (Google Drive → parse → Sheets → memory)
  ├── generates health report with leisure/relaxation recommendations
  └── delivers to Telegram

22:00 agentTurn (isolated) — Next-Day Plan
  ├── reads health data + health report leisure recommendations
  ├── generates next-day plan
  ├── delivers plan to Telegram
  ├── writes plan to {PLAN_FILE}
  └── runs update-reminders.sh → updates system crontab

22:05 system crontab → parse-plan-update-reminders.sh (fallback)

Next day: system crontab → tg-remind.sh → curl Telegram Bot API (zero LLM)

Setup

1. Install scripts

Copy scripts/ to a stable path (e.g. ~/.openclaw/scripts/):

• tg-remind.sh — edit BOT_TOKEN and CHAT_ID
• update-reminders.sh — edit REMIND path
• parse-plan-update-reminders.sh — edit PLAN_FILE and UPDATE_SCRIPT paths
• sleep_import.py — edit ACCOUNT, DRIVE_FOLDER_ID, SHEET_ID constants

2. Prerequisites for sleep_import.py

• gog CLI installed and authenticated (gog auth login)
• Google Drive folder with Health Auto Export JSON files
• Google Sheets for sleep log
• Local workspace dirs: tmp/sleep-import/, memory/

3. System crontab

# Fallback plan parser (22:05 local, convert to UTC)
5 <UTC_HOUR> * * * /path/to/parse-plan-update-reminders.sh

4. OpenClaw cron jobs

15:00 Health Report:

• schedule: 0 15 * * * (your timezone)
• sessionTarget: isolated
• payload: agentTurn — run sleep_import.py then generate report per references/health-report.md
• timeoutSeconds: 600
• delivery: telegram

22:00 Next-Day Plan:

• schedule: 0 22 * * * (your timezone)
• sessionTarget: isolated
• payload: agentTurn — generate plan per references/briefing-prompt.md, write plan file
• timeoutSeconds: 300
• delivery: telegram

5. Configure reference files

Edit both files in references/ — replace all {PLACEHOLDER} values with actual paths.

Scripts

sleep_import.py

Full pipeline: searches Google Drive for HealthAutoExport-YYYY-MM-DD.json (last 3 days), downloads, parses Apple Health metrics (sleep, HR, HRV, SpO2, respiratory rate, wrist temp, breathing disturbances), appends to Google Sheets, and updates local memory file.

python3 scripts/sleep_import.py

Output: human-readable sleep report with metrics, assessment, and advice.

update-reminders.sh

Flags: --recovery, --training, --english, --openclaw1..3 — all HH:MM local time. Only pass that day's events. OpenClaw = start time (auto -10min for warning).

Reference Files

• references/briefing-prompt.md — Complete 22:00 plan generation rules. See for: life-day rules, schedule constraints, health report integration, self-check checklist.
• references/health-report.md — Complete 15:00 health report rules. See for: data pipeline, report format, leisure recommendation format.

Design Decisions

• Two isolated cron jobs: health report (15:00) and plan (22:00) run independently. Plan reads health report output, not the reverse.
• System crontab for reminders: LLM-based reminders unreliable. curl Telegram is mechanical and always works.
• Isolated sessions: 3x cheaper than session-bound (~$0.9 vs ~$2.6/run), stable context size.
• Plan file decoupling: 22:05 fallback parser catches cases where agent didn't update crontab.
• Health report → plan pipeline: health report includes leisure recommendations that the plan generator must incorporate. This creates a data dependency, not a runtime dependency.