ClawCoach Food — Photo Analysis & Meal Logging

This skill handles food photo analysis via Claude Vision, text-based meal logging, and the confirmation flow.

When to Activate

• User sends a photo — assume it is food unless context clearly suggests otherwise
• User types a food description ("I had 2 eggs and toast for breakfast")
• User says "log [food]" or "I ate [food]"
• User wants to edit or delete a previous meal

Data Storage

All meals are stored in ~/.clawcoach/food-log.json with this structure:

{
  "meals": [
    {
      "id": "2026-02-22-lunch-001",
      "date": "2026-02-22",
      "type": "lunch",
      "status": "confirmed",
      "items": [
        {
          "name": "grilled chicken breast",
          "portion": "6 oz",
          "calories": 280,
          "protein_g": 52,
          "fat_g": 6,
          "carbs_g": 0
        }
      ],
      "total_calories": 520,
      "total_protein_g": 62,
      "total_fat_g": 14,
      "total_carbs_g": 48,
      "source": "photo",
      "timestamp": "2026-02-22T12:35:00Z"
    }
  ]
}

Photo Analysis Flow

When the user sends a photo:

1. Analyze the image using your vision capabilities. Identify every distinct food item visible. For each item estimate:
• Name (be specific: "grilled chicken breast" not just "chicken")
• Portion in common units (oz, cups, pieces, slices)
• Calories and macros (protein, fat, carbs in grams)

Use your nutritional knowledge. For common foods, these are well-established values. Be conservative with portions if uncertain.

1. Present the results in the user's persona voice:
• List each item with portion and macros
• Show meal total
• Show daily running totals (consumed / target / remaining)
• Ask: "confirm? (yes / edit / redo)"

1. Handle response:
• "yes" / "confirm" — Write the meal to ~/.clawcoach/food-log.json with status "confirmed"
• Correction (e.g., "the rice was brown rice" or "it was more like 8oz") — recalculate and present updated totals
• "redo" — ask for a new photo or text description

1. After confirmation, always show updated daily totals.

Text-Based Logging

When the user describes food in text:

1. Parse the food items and estimate portions from the description
2. Calculate macros for each item using your nutritional knowledge
3. Follow the same confirmation flow as photo analysis

Meal Type Auto-Detection

Categorize meals by time:

• Before 10:00 = breakfast
• 10:00 - 14:00 = lunch
• 14:00 - 17:00 = snack
• After 17:00 = dinner

The user can override: "log this as a snack"

Editing and Deleting

• "Delete my lunch" — find today's lunch entry, remove it from food-log.json
• "I think that was more like 400 calories" — update the specific meal entry
• "What did I eat today?" — list all confirmed meals for today with totals

Daily Totals

After any meal is confirmed, calculate and show:

1. Read profile from ~/.clawcoach/profile.json for targets
2. Sum all confirmed meals for today from food-log.json
3. Display:
• Consumed: X cal | Xg protein | Xg fat | Xg carbs
• Target: X cal | Xg protein | Xg fat | Xg carbs
• Remaining: X cal | Xg protein | Xg fat | Xg carbs

Edge Cases

• Blurry or unclear photo: "I can't quite make out the food. Try a better lit photo, or just tell me what you had."
• Non-food photo: "That doesn't look like food! Send a photo of your meal, or type what you ate."
• Unknown food: Ask the user for clarification rather than guessing wildly.
• Multiple items unclear: "I can see chicken and something else — is that rice or pasta?"
• No portion visible: Use standard serving sizes and note: "I estimated a standard portion — let me know if it was more or less."

Nutritional Reference (Common Foods per 100g)

Use these as a baseline. Scale by estimated portion size.

For foods not on this list, use your general nutritional knowledge. Be transparent when estimating.

Important

• Always present macros rounded to whole numbers
• Always show daily running totals after confirming a meal
• The persona voice comes from clawcoach-core — match it in all responses
• Never log a meal without user confirmation
• Generate unique meal IDs as: {date}-{meal_type}-{sequence}