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:

Rule	Requirement	Why
Float amounts	1.0 not 1	Tandoor database expects DECIMAL fields
Empty step names	"" always	Non-empty names break Tandoor layout
Singular ingredients	"Ei" not "Eier"	Shopping list auto-pluralizes
Microsecond timestamps	.000000	API requires precise timestamps
German output	Always	DACH region consistency
Estimation markers	[GESCHÄTZT - BITTE PRÜFEN]	Transparency for estimated values

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

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

   Score	Condition	Action
   ≥80	Clear text	Process normally
   60-79	Slightly unclear	Process with warnings
   40-59	Significant issues	Extract what's possible, mark estimations
   <40	Illegible	Return structured error with partial data

3. 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

2. Unit Conversion:

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

     Original	Target	Factor
     cup (liquid)	ml	240
     cup (flour)	g	120
     tbsp/EL	ml	15
     tsp/TL	ml	5
     oz (weight)	g	30
     lb	g	450
     stick butter	g	115

3. Ingredient Normalization:

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

     Input	Standard German
     eggs, Eier	Ei
     onions, Zwiebeln	Zwiebel
     carrots, Möhren	Karotte
     cream, Sahne	Sahne

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

4. Plural Exceptions: Some items remain plural:

   • Spaghetti → "always_use_plural_food": true
   • Pommes frites → "always_use_plural_food": 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]"

2. Time Estimation (if missing):

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

     Dish Type	Working	Waiting
     Salat	15	0
     Pasta (einfach)	15	15
     Kuchen (Rührkuchen)	20	45
     Eintopf	30	90
     Default	20	30

   • Mark with [GESCHÄTZT - BITTE PRÜFEN]

3. Servings (if missing):

   • Standard: 4 servings
   • Mark as estimated

4. 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:

   Test	Check	On Fail
   JSON syntax	JSON.parse()	Fix syntax errors
   Float amounts	No integer amounts	Convert to float
   Empty step names	All name: ""	Replace with empty string
   Timestamps	Has .000000	Add microseconds
   Singular ingredients	No plurals (except exceptions)	Convert to singular

2. Quality Score Calculation:

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

   Score	Status	Action
   ≥95	✅ Ready	Import directly
   ≥80	⚠️ Warnings	Review warnings first
   ≥60	🔍 Review	Manual verification needed
   <60	❌ Reject	Needs rework

3. Plausibility Checks:

   • working_time: 5-300 minutes
   • waiting_time: 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`

```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