smart-landscape-designer

name	Smart Landscape Designer
slug	smart-landscape-designer
description	Generate copy-ready landscape image prompts from user requests and optional reference images.
category	image_prompt_generation
execution_mode	llm-only
chainTo	image-creator
icon	image
version	1.3.0
author	SmartAIHub
isAutoTrigger	false
enabledByDefault	true
priority	55
creditMultiplier	1

Smart Landscape Designer Skill V1.3.0

This is a vendor-neutral agent skill for generating one copy-ready landscape image prompt as a single raw string.

It is designed to work across Gemini, OpenAI, Claude, and OpenCode style runtimes because it uses:

• Markdown instructions
• JSON input/output schemas
• A UI schema with explicit field mapping
• No vendor-specific tool calling

Package contents

• skill.md
• system_prompt.md
• knowledge/landscape_knowledge_base.md
• knowledge/quick_start_guide.md
• schemas/input.schema.json
• schemas/output.schema.json
• schemas/ui.schema.json
• review_report.md

What changed in V1.3.0

• Simplified reference images — reference_images is now a flat array of image URLs (strings) instead of an array of objects with role/notes
• New ImageSourcePicker — users can pick images from Upload, Library (personal/shared/group), or paste URLs
• Removed hidden role field — every image is always treated as reference; no need to show a read-only field
• Added referenceNotes — a single optional textarea for notes about all reference images (mention images by number)
• Retains Output Language, Mode Override, and the clearer variation controls

Input contract

The runtime must send JSON matching schemas/input.schema.json.

Key inputs:

• userRequest: the landscape change or creation request
• outputLanguage: the language of the final returned prompt (en or th)
• reference_images: flat array of image URLs (strings) — all treated as visual references
• referenceNotes: optional free-text notes about how to use the reference images
• modeOverride: defaults to mode5_text_to_landscape
• variationType: controls how the skill chooses one final prompt
• numberOfOptions: how many internal candidates may be drafted before selecting the final prompt
• customSelection: clear preset creative directions for the final prompt
• customSelectionNotes: optional extra direction
• maxPromptChars: hard limit for the final single prompt

All input fields, including nested constraint fields, have explicit defaults.

Attachment handling

Reference images are selected via the ImageSourcePicker UI (upload, library, or URL). The JSON payload stores image URLs in the reference_images array. The referenceNotes field provides optional context about how to use each image.

Output contract

Return output matching schemas/output.schema.json.

The output must be:

• a single raw string
• copy-ready for an image model
• free of JSON wrappers, arrays, labels, bullet lists, and code fences

Runtime behavior

1. Ignore missing image attachments.
2. Detect the best mode unless modeOverride is supplied.
3. When reference images are attached for an existing-property request, preserve the building unless disabled.
4. Build photo-real prompts ready to paste into an image model.
5. If variationType uses internal alternatives, draft up to numberOfOptions candidates, then return only the best single prompt.
6. Use customSelection and customSelectionNotes when the custom-guided path is chosen.
7. Keep the final prompt under maxPromptChars.
8. Return the prompt in the requested outputLanguage.

Single-string rule

Do not produce:

• JSON objects
• arrays
• menus
• tutorials
• bullet explanations
• conversational wrappers such as "Here is your prompt"

Self-review requirements

Before returning:

1. Check that the output schema is satisfied.
2. Check that the output is a single string only.
3. Check that reference_images URLs and referenceNotes are considered when present.
4. Check that preservation wording appears when required.
5. Check that every input field has a default.