Preparing Plant Data via AI Prompt¶

Kamerplanter manages 80+ structured fields per plant species -- from taxonomy and nutrient profiles to pest data and companion planting relationships. Manually collecting this data from various sources is time-consuming. That's why we use Claude Code Agents to fully prepare and quality-check new plants.

Overview: The AI Pipeline¶

flowchart LR
    A["Plant name\n(e.g. 'Basil')"] --> B["plant-info-document-generator"]
    B --> C["Plant document\nspec/ref/plant-info/*.md"]
    C --> D["agrobiology-requirements-reviewer"]
    D --> E["Review report\nspec/requirements-analysis/"]
    E -->|Corrections| C
    C --> F["Seed data / CSV import\n(REQ-012)"]

The workflow consists of three steps:

1. Generation -- An AI agent researches and creates the plant document
2. Review -- A second agent checks the data for scientific accuracy
3. Import -- The reviewed document serves as the basis for data import

Step 1: Generate Plant Document¶

The plant-info-document-generator agent automatically researches all relevant data and creates a structured Markdown document under spec/ref/plant-info/.

Invocation in Claude Code¶

Create a plant document for basil

or for multiple plants at once:

Create plant documents for: rosemary, thyme, oregano, sage

Claude Code recognizes the context and automatically activates the plant-info-document-generator agent.

What the Agent Does¶

1. Analyze input -- Identifies the scientific name, family, and genus
2. Research -- Searches the web for:
   • Taxonomy and master data (GBIF, RHS, USDA)
   • Growth phases with PPFD, VPD, temperature per phase
   • Nutrient profiles (NPK, EC, pH per phase)
   • Pests and diseases with beneficial organisms
   • Care and overwintering instructions
   • Crop rotation and companion planting partners
3. Create document -- Writes a complete document with all Kamerplanter field references

Result¶

The document is saved as:

spec/ref/plant-info/<scientific_name_snake_case>.md

Example: spec/ref/plant-info/ocimum_basilicum.md

Document Structure¶

Each generated document contains these sections:

Each table includes a KA-Field column referencing the exact Kamerplanter database field.

Step 2: Scientific Review¶

The agrobiology-requirements-reviewer agent checks the document from an agrobiology expert's perspective.

Invocation in Claude Code¶

Review the plant document spec/ref/plant-info/ocimum_basilicum.md for scientific accuracy

What the Review Agent Checks¶

• Taxonomy -- Scientific names per APG IV, correct family assignment
• Light data -- PPFD/DLI instead of Lux, photoperiodism correct
• Climate data -- VPD calculation plausible, day/night temperature separated
• Nutrients -- EC ranges realistic, mixing order correct (CalMag before sulfates)
• Pests -- Scientific names, IPM tier approach (prevention > monitoring > intervention)
• Toxicity -- ASPCA data for cats/dogs verified
• Companion planting -- Compatibilities biologically justified

Result¶

The review report is saved under:

spec/requirements-analysis/plant-info-agrobiology-review-<batch>.md

Findings are classified as:

Category	Meaning
Scientifically Wrong	Immediate correction needed
Incomplete	Important aspects missing
Too Vague	Needs precision
Note	Best practice recommendation

Step 3: Import into Kamerplanter¶

The reviewed documents serve as the basis for data import:

Option A: Seed Data (Developer)¶

Plant data is built into a Python seed script at src/backend/app/migrations/seed_plant_info.py. The seed reads the Markdown documents and creates the corresponding ArangoDB documents.

Option B: CSV Import (End User)¶

Each plant document contains ready-made CSV lines in section 8 that can be uploaded via the REQ-012 import function.

Existing Plant Documents¶

Currently 32 plants are fully documented:

spec/ref/plant-info/

Including vegetables (tomato, pepper, cucumber, zucchini, ...), herbs (basil, parsley, dill, chives, ...), ornamentals (dahlia, petunia, sunflower, ...), and houseplants (monstera, peace lily, spider plant, guzmania).

Tips for Best Results¶

Involved Claude Code Agents¶