Llmn

Meal Planning Wizard

This skill guides users through creating an optimized meal plan using the llmn CLI tool. It uses quadratic programming to find diverse food combinations that satisfy nutritional constraints.

Optimization Modes

Choose the right mode based on user needs:

Template-based is recommended for most users because it produces meals that look like what humans actually eat (1 protein + 1 legume + vegetables per meal).

Workflow

Step 0: Check for Existing User Profile

Before gathering goals, check if the user has an existing profile:

uv run llmn user show --json

If profile exists with goal set (e.g., goal: "fat_loss:185:165"):

• Extract current weight and target weight from the goal
• Offer to use stored targets: "I see you have a profile for fat loss (185 lbs → 165 lbs). Should I use those targets?"
• If user agrees, skip to Step 2 and use --use-profile flag in Step 4
• If user wants different targets, continue to Step 1

If profile exists without goal:

• Use profile's age, sex, height, weight, activity for calculations
• Ask only about goal type and calorie/protein targets

If no profile exists:

• Continue to Step 1 to gather all information
• After optimization, suggest creating a profile: llmn user create

Step 1: Gather User Goals

IMPORTANT: Always confirm calorie and protein targets with the user BEFORE running any optimization.

Ask the user about their dietary goals. Use the AskUserQuestion tool to collect:

1. Diet type (one of):

   • Omnivore (eats everything)
   • Pescatarian (fish + eggs, no meat)
   • Vegetarian (eggs + dairy, no meat/fish)
   • Vegan (plant-based only)

2. Diet style (one of):

   • Standard (balanced macros)
   • Slow-carb (legumes as carbs, no white carbs/fruit)
   • Low-carb (reduced carbohydrates)
   • High-protein (muscle building focus)
   • Mediterranean (olive oil, fish, vegetables)

3. Goal (one of):

   • Weight loss / Cutting (1600-1800 cal, 150g protein)
   • Muscle gain / Bulking (2800-3200 cal, 180g protein)
   • Maintenance (2200-2400 cal, 120g protein)
   • Custom (user specifies)

4. Calorie range: Get min/max daily calories

5. Protein target: Get minimum daily protein in grams

Step 2: Check Available Foods

Search for foods that match the user's diet type:

# Check what staple foods are already tagged
uv run llmn tags list --tag staple --json

If few staples are tagged, suggest running the tagging workflow or offer to tag common foods for their diet type.

Step 3: Create Constraint Profile

Create a YAML profile file at /tmp/llmn_profile.yaml:

name: user_custom_plan
description: "Custom plan - [diet_type] [diet_style] [goal]"

calories:
  min: <cal_min>
  max: <cal_max>

nutrients:
  protein:
    min: <protein_min>
  fiber:
    min: 30
  sodium:
    max: 2300

include_tags:
  - staple

exclude_tags:
  - exclude

options:
  max_grams_per_food: 500

For low-carb diets, add:

  carbohydrate:
    max: 100

Step 4: Run Optimization

Recommended: Template-based optimization (produces realistic meals):

# Map diet type to pattern
# - Pescatarian → --pattern pescatarian
# - Vegetarian → --pattern vegetarian
# - Vegan → --pattern vegan
# - Omnivore → --pattern mediterranean (or paleo)

# Map diet style to additional pattern
# - Slow-carb → --pattern slow_carb
# - Low-carb/Keto → --pattern keto

If user has a profile with goal set (from Step 0):

# Use --use-profile to auto-derive calorie/protein targets from stored user data
uv run llmn optimize --pattern pescatarian --template --use-profile --json

If no profile or user wants custom targets:

# Use explicit --goal flag with weight info
uv run llmn optimize --pattern pescatarian --template --goal "fat_loss:185lbs:165lbs" --json

Alternative: YAML profile-based optimization (most control):

uv run llmn optimize --file /tmp/llmn_profile.yaml --json

Step 5: Present Results

Parse the JSON output and present to the user:

• List of foods with amounts in grams
• Total calories and protein achieved
• Any binding constraints (at their limits)
• Suggestions from the optimizer

If optimization fails (infeasible), explain which constraints conflict and suggest relaxations.

Step 6: Iterate

Ask if the user wants to:

• Adjust constraints and re-run
• Exclude specific foods they don't like
• Add more foods to their staples
• Generate a recipe prompt for Claude

Key Commands Reference

# Search for foods
uv run llmn search "chicken" --json

# Get food nutrient details
uv run llmn info <fdc_id> --json

# Tag a food as staple
uv run llmn tags add <fdc_id> staple

# Template-based optimization (recommended)
uv run llmn optimize --pattern pescatarian --template --json
uv run llmn optimize --pattern pescatarian --pattern slow_carb --template --seed 42 --json

# Profile-based optimization
uv run llmn optimize --file /tmp/profile.yaml --json

# What-if analysis
uv run llmn explore whatif --base latest --add "protein:min:200" --json

# Export for recipe generation
uv run llmn export-for-llm latest

Available Patterns

Patterns can be combined: --pattern pescatarian --pattern slow_carb

Common Food IDs for Quick Tagging

Proteins

• 175167: Salmon (Atlantic, raw)
• 175159: Tuna (bluefin, raw)
• 171287: Eggs (whole, raw)
• 171077: Chicken breast (raw)
• 172421: Lentils (cooked)
• 173757: Chickpeas (cooked)

Vegetables

• 168462: Spinach (raw)
• 169967: Broccoli (raw)
• 168421: Kale (raw)
• 170108: Bell peppers (raw)

Fats

• 171705: Avocado (raw)
• 748608: Olive oil
• 170567: Almonds (raw)

Example Session

User: "I want to lose weight, I'm pescatarian"

1. Ask: calorie target? (suggest 1600-1800 for weight loss)
2. Ask: protein target? (suggest 150g for satiety)
3. Run template-based optimization:

   uv run llmn optimize --pattern pescatarian --pattern slow_carb --template --goal "fat_loss:185lbs:165lbs" --json
   

4. Show results with per-meal breakdown:
   • Breakfast: Eggs 150g, Edamame 100g, Kale 100g
   • Lunch: Cod 235g, Kidney beans 216g, Zucchini 161g
   • Dinner: Salmon 170g, Lentils 113g, Carrots 149g
   • Snack: Peanuts 30g
5. Offer to: adjust constraints, exclude foods, or generate recipes