Diet Record

Fields:

• photo_auto_log: true = auto-log on photo upload, false = confirm first, null = not yet set.
• dietary_restrictions: e.g. ["素食", "清真", "无麸质", "低碳水"]
• allergies: e.g. ["花生", "海鲜", "乳糖不耐"]
• disliked_foods: foods user explicitly dislikes
• favorite_foods: frequently eaten or preferred foods
• diet_goal: e.g. "减脂", "增肌", "维持体重", "均衡饮食"
• daily_calorie_target: e.g. 1800 (kcal), null if not set
• meal_routine: e.g. "一日三餐", "16:8轻断食", "少食多餐"
• notes: any other dietary habits or notes from user

Preference Discovery

Photo auto-log preference: On the first food photo upload (or when photo_auto_log is null), recognize items as usual, then ask: "以后发食物照片时，要自动帮你记录饮食吗？还是每次先确认再记录？"

Dietary habits: Whenever user mentions dietary preferences, restrictions, allergies, goals, or habits in conversation, extract and save to the corresponding fields. Examples:

• "我对花生过敏" → add "花生" to allergies
• "我在减脂" → set diet_goal to "减脂"
• "我不吃香菜" → add "香菜" to disliked_foods
• "我每天控制在1500卡" → set daily_calorie_target to 1500
• "我在做16:8轻断食" → set meal_routine to "16:8轻断食"

Preferences are accumulated over time — update individual fields without overwriting unrelated ones. Read preferences before each interaction to provide personalized feedback (e.g. warn if a meal exceeds calorie target, flag allergens in recognized food).

Workflow

Photo Input

1. Receive food photo from user
2. Read diet-preferences.json to check photo_auto_log
3. Analyze the image: identify each food item, estimate portion size
4. Look up nutrition data per item (use the reference table below)
5. If photo_auto_log is null: present result, ask preference (see above), then log
6. If photo_auto_log is true: calculate totals, log directly, respond with summary
7. If photo_auto_log is false: present recognized items — ask user to confirm or correct, then log
8. Append record to diet-log.jsonl

Text Input

1. Parse food description (e.g. "一碗牛肉面加个煎蛋")
2. Break into individual items with estimated portions
3. Look up nutrition and calculate totals
4. Append record

Daily Summary

When user asks "今天吃了什么" or "daily summary":

python3 -c "
import json, datetime
today = datetime.date.today().isoformat()
with open('diet-log.jsonl') as f:
    meals = [json.loads(l) for l in f if today in l]
total = sum(m['total_calories'] for m in meals)
print(f'Today: {len(meals)} meals, {total:.0f} kcal')
for m in meals:
    items = ', '.join(i['name'] for i in m['items'])
    print(f\"  {m['meal_type']}: {m['total_calories']:.0f} kcal — {items}\")
"

Nutrition Reference (per 100g)

Common foods for quick lookup (no API needed):

For unlisted foods, estimate based on similar items or ask the user for specifics.

Cooking Method Calorie Adjustments

• 清蒸/水煮: baseline
• 炒(少油): +10-15%
• 炒(多油): +20-30%
• 油炸: +30-50%
• 烧烤: -5% (fat drips off)

Portion Estimation

• 一碗米饭 ≈ 200g
• 一盘菜 ≈ 200-300g
• 一个拳头 ≈ 150g (水果/肉)
• 一汤匙油 ≈ 10g (90 kcal)
• 一杯(240ml)牛奶 ≈ 245g

Key Rules

• Present results in a clean table format
• When recognition confidence is low, ask user to confirm before logging
• Auto-detect meal_type from time: 06-10 breakfast, 11-14 lunch, 17-20 dinner, else snack
• Remind if daily protein is under 1.2g/kg body weight (when user weight is known)
• Never make moral judgments about food choices