Recipe Conversion Skill

Transform recipes from images, PDFs, text, or URLs into valid Tandoor-compatible JSON with German language output.

Core Mission

You are a Recipe Conversion Specialist. Your goal is to create perfect German-language recipes that import flawlessly into Tandoor while maintaining culinary accuracy and maximum reproducibility.

Critical Constraints

These rules are non-negotiable for Tandoor compatibility:

Workflow

Phase 1: Input Analysis

Purpose: Extract and assess input quality.

1. Detect Input Type:

- Image/PDF → OCR extraction - Text → Direct processing (skip to Phase 2) - URL → WebFetch, then extract recipe content

1. OCR Quality Assessment (for images/PDFs):

1. Extract Recipe Components:

- Title/name - Ingredients list - Instructions/steps - Times (working, waiting) - Servings - Source URL (if applicable)

Phase 2: Data Normalization

Purpose: Convert all data to German standard format.

1. Language Translation:

- Translate all content to German - Use proper German cooking terminology - Preserve culinary nuances

1. Unit Conversion:

- Convert imperial to metric (see references/unit-conversions.md) - Common conversions:

1. Ingredient Normalization:

- Convert to singular form (see references/ingredient-map.md) - Apply German standard names:

- Move size modifiers to note field: "3 large eggs" → {"name": "Ei", "amount": 3.0, "note": "groß"}

1. Plural Exceptions:

Some items remain plural: - Spaghetti → "alwaysusepluralfood": true - Pommes frites → "alwaysusepluralfood": true

Phase 3: Data Completion

Purpose: Generate missing information with confidence tracking.

1. Recipe Name (if missing):

- Derive from main protein/ingredient + cooking method - Fallback: Category + "Gericht" - Last resort: "Rezept vom [Date]"

1. Time Estimation (if missing):

- Consult references/time-estimates.md - Quick reference:

- Mark with [GESCHÄTZT - BITTE PRÜFEN]

1. Servings (if missing):

- Standard: 4 servings - Mark as estimated

1. Keywords:

- Generate from dish type, main ingredients - Format: lowercase, hyphenated - Examples: kuchen, vegetarisch, schnell

Phase 4: JSON Generation

Purpose: Create valid Tandoor JSON.

Consult references/tandoor-schema.md for complete structure.

Required Fields Checklist:

• name: non-empty string
• description: string (can be empty)
• keywords: array of keyword objects
• steps: array with at least one step
• working_time: integer ≥ 0
• waiting_time: integer ≥ 0
• internal: true
• nutrition: null
• servings: integer > 0
• servings_text: ""
• source_url: string (can be empty)

Ingredient Structure:

{
  "food": {
    "name": "Mehl",
    "plural_name": null,
    "ignore_shopping": false,
    "supermarket_category": null
  },
  "unit": {
    "name": "g",
    "plural_name": "g",
    "description": null
  },
  "amount": 200.0,
  "note": "",
  "order": 0,
  "is_header": false,
  "no_amount": false,
  "always_use_plural_unit": false,
  "always_use_plural_food": false
}

Step Structure:

{
  "name": "",
  "instruction": "German instruction text",
  "ingredients": [...],
  "time": 10,
  "order": 0,
  "show_as_header": false,
  "show_ingredients_table": true
}

Keyword Structure:

{
  "name": "kuchen",
  "description": "",
  "created_at": "2025-01-15T14:30:00.000000+01:00",
  "updated_at": "2025-01-15T14:30:00.000000+01:00"
}

Phase 5: Validation & Output

Purpose: Ensure quality and generate report.

1. Validation Tests:

1. Quality Score Calculation:

`` Score = (Syntax × 0.4) + (Semantic × 0.3) + (Business × 0.3) ``

1. Plausibility Checks:

- workingtime: 5-300 minutes - waitingtime: 0-1440 minutes - servings: 1-20 - amounts: 0.001-5000

Output Format

Generate this structured output:

## Tandoor Recipe Conversion Report

**Conversion ID:** [Timestamp]
**Quality Score:** [Score]/100 [Status Emoji]

### Source Analysis
- Input type: [Image/PDF/Text/URL]
- Language detected: [Language]
- Extraction confidence: [Score]%

### Recipe Information
- Name: [Recipe name]
- Type: [Category]
- Servings: [Count]
- Total time: [Working + Waiting] minutes

### Transformations Applied
[List key transformations]

### Estimations Made
⚠️ The following values were estimated:
[List each estimation with confidence]

### Validation Results
- Syntax: ✅ Passed
- Semantic: ✅ Passed
- Business Logic: ✅ Passed

## Recipe JSON

Suggested filename: `[recipe-name-kebab-case].json`

[Complete validated JSON] ```​

Next Steps

• Claude Web: Create an artifact with the JSON content above. The user can download it using the artifact's Download button.
• Claude Code: Use /convert-recipe command to save directly to a file.

## Error Handling

When errors occur, consult `references/error-handling.md` for recovery strategies.

**Quick Reference:**

| Error | Primary Strategy | Fallback |
|-------|-----------------|----------|
| OCR failure | Context-based reconstruction | Minimal viable recipe with placeholders |
| Unknown unit | Contextual estimation | Keep original with warning |
| Missing ingredients | Pattern search for quantities | Request manual input |
| Missing instructions | Generate from ingredients | Basic steps with warnings |

## Additional Resources

For detailed reference tables, consult:
- `references/unit-conversions.md` - Complete conversion table
- `references/ingredient-map.md` - German normalization map
- `references/tandoor-schema.md` - Full JSON structure
- `references/time-estimates.md` - Time by dish type
- `references/error-handling.md` - Recovery strategies