FoodLens

AI-powered meal photo recognition and nutrition tracking. Use when a user sends a food/meal photo with keywords like breakfast, lunch, dinner, snack, or "wha...

Security Scan

The skill's requirements and instructions are coherent for a local meal-photo nutrition tracker, but it executes local Python code and sends images to LLM vision tools — review local files and privacy implications before use.

Purpose & Capability

Name/description (meal-photo recognition, nutrition tracking) match the actions described: saving photos, running analyze_photo.py, consulting a local nutrition_db, appending JSON logs, and producing summaries. No unrelated credentials, binaries, or services are requested.

Instruction Scope

Instructions explicitly instruct the agent to save photos, activate a virtualenv, run local Python scripts (analyze_photo.py) and import from a local module (foodlens), then write JSON logs under a user directory. This is expected for a local tracker, but it means the skill will execute arbitrary Python code present in $FOODLENS_DIR and will read/write files there. Also the fallback uses an 'image' tool / GPT-4o Vision which will send image data to external model providers.

Install Mechanism

No install spec or third-party downloads are present (instruction-only). That reduces supply-chain risk, but it requires the user to supply the code, venv, and dependencies. The skill assumes local project files exist; nothing is downloaded automatically.

Credentials

The skill declares no required environment variables, credentials, or config paths beyond optional defaults for $FOODLENS_DIR and venv. The only notable external interaction is use of LLM vision/image tools (model/provider usage), which is consistent with image analysis but has privacy implications.

Persistence & Privilege

always:false (no forced global inclusion). The skill writes persistent data (data/YYYY-MM-DD.json) and expects a venv and local files under the specified workspace directory. Writing user-local logs is coherent for the purpose but grants the skill filesystem persistence within that directory.

Assessment

This skill appears to do what it says, but it runs local Python code and sends images to LLM vision tools. Before installing or invoking: (1) Inspect the contents of $FOODLENS_DIR (especially analyze_photo.py, nutrition_db.py, and any activation scripts) to ensure they are safe and from a trusted source; (2) Run the code in an isolated environment (dedicated venv or container); (3) Be aware that images will be sent to external model providers (privacy risk) — avoid sending sensitive photos; (4) Back up any data you care about before letting the skill write to data/YYYY-MM-DD.json; (5) If you don’t already have the project code, obtain it from a trustworthy source rather than relying on this instruction-only skill to function automatically.

FoodLens — AI Meal Photo & Nutrition Tracker

Trigger Conditions

User sends a meal/food photo with context such as:

• breakfast / lunch / dinner / snack / supper / 早饭 / 午饭 / 晚饭 / 加餐 / 零食

Configuration

Set these paths for your deployment (defaults shown):

FOODLENS_DIR=~/.openclaw/workspace/skills/foodlens
FOODLENS_DATA=$FOODLENS_DIR/data          # daily JSON logs: YYYY-MM-DD.json
FOODLENS_VENV=$FOODLENS_DIR/venv

Nutrition goals are user-configurable. Defaults (edit foodlens_config.json):

• Calories: 2000 kcal/day
• Protein: 80g | Carbs: 250g | Fat: 65g

---

Core Flow

Step 1 — Analyze Photo (Primary)

Save the inbound photo to a temp path, then run:

cd $FOODLENS_DIR && source venv/bin/activate
python3 analyze_photo.py /path/to/photo.jpg "lunch"

This script:

1. Calls GPT-4o Vision (fallback: Gemini) to identify foods and estimate portion sizes using container/utensil references
2. Cross-validates against nutrition_db (778 foods + 197 aliases); if deviation > 30%, trusts the database
3. Appends the meal to data/YYYY-MM-DD.json
4. Outputs a formatted nutrition report

Forward the script output directly to the user.

---

Step 2 — Fallback (API unavailable)

If analyze_photo.py fails, use the image tool:

image(
  image="/path/to/photo.jpg",
  prompt="You are a professional nutritionist. Identify all foods in this meal
  photo. Observe container size and utensils to estimate actual grams per item.
  Reference: standard takeout box 500–800 ml, bowl of rice ~150–200 g,
  stir-fried noodles ~400–500 g. List each food: name, estimated grams,
  kcal per 100 g, protein/carb/fat per 100 g."
)

Then write results via Python:

cd $FOODLENS_DIR && source venv/bin/activate && python3 - <<'EOF'
import json, uuid, sys
sys.path.insert(0, '.')
from foodlens import (ensure_item_nutrition, calc_total,
                      health_score_and_comment, load_day, save_day,
                      today_str, recalc_day_totals)
from datetime import datetime

date_str = today_str()
day = load_day(date_str)

# Replace with image tool results
items = [
    ensure_item_nutrition({'name': 'food name', 'grams': 300, 'source': 'image_tool'}),
]

meal_total = calc_total(items)
score, comment = health_score_and_comment(meal_total, len(items))
meal = {
    'meal_id': f'meal_{uuid.uuid4().hex[:10]}',
    'timestamp': datetime.now().isoformat(),
    'label': 'lunch',
    'items': items,
    'meal_total': meal_total,
    'health_score': score,
    'comment': comment,
}
day['meals'].append(meal)
recalc_day_totals(day)
save_day(date_str, day)
print(json.dumps({'meal': meal, 'daily_total': day['daily_total']}, ensure_ascii=False, indent=2))
EOF

---

Step 3 — Format Reply

🍽️ [Lunch] Nutrition Analysis

🔍 Identified foods:
  • Stir-fried noodles ~400g (720 kcal)
  • Shrimp ~30g (27 kcal)
  • Chicken slices ~60g (90 kcal)

📊 Meal total:
  • Calories: 837 kcal
  • Protein: 38g | Carbs: 102g | Fat: 29g

⭐ Health score: 7/10
  Comment: ...

📈 Daily total (meal N):
  • Calories: X / [goal] kcal (X%)
  • Protein: X / [goal]g (X%)

---

Step 4 — User Corrections

If user says "that's not X it's Y" or "only about Xg":

1. Re-query nutrition_db for the corrected food
2. Update the JSON entry
3. Reply with corrected nutrition totals

---

Step 5 — Duplicate Detection

If the same photo is sent again, alert the user it was already logged and ask whether to record again.

---

Summaries

Daily summary:

cd $FOODLENS_DIR && source venv/bin/activate
python3 analyze_photo.py --summary today

Weekly trend (last 7 days):

cd $FOODLENS_DIR && source venv/bin/activate
python3 analyze_photo.py --weekly-summary yesterday 7

---

Data Layout

Path	Description
data/YYYY-MM-DD.json	Daily meal logs
nutrition_db.py	778 foods + 197 aliases
analyze_photo.py	Main entry point
foodlens_config.json	User nutrition goals
venv/	Python virtual environment