| 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.mdsystem_prompt.mdknowledge/landscape_knowledge_base.mdknowledge/quick_start_guide.mdschemas/input.schema.jsonschemas/output.schema.jsonschemas/ui.schema.jsonreview_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 requestoutputLanguage: the language of the final returned prompt (en or th)reference_images: flat array of image URLs (strings) — all treated as visual referencesreferenceNotes: optional free-text notes about how to use the reference imagesmodeOverride: defaults to mode5_text_to_landscapevariationType: controls how the skill chooses one final promptnumberOfOptions: how many internal candidates may be drafted before selecting the final promptcustomSelection: clear preset creative directions for the final promptcustomSelectionNotes: optional extra directionmaxPromptChars: 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
- Ignore missing image attachments.
- Detect the best mode unless
modeOverride is supplied.
- When reference images are attached for an existing-property request, preserve the building unless disabled.
- Build photo-real prompts ready to paste into an image model.
- If
variationType uses internal alternatives, draft up to numberOfOptions candidates, then return only the best single prompt.
- Use
customSelection and customSelectionNotes when the custom-guided path is chosen.
- Keep the final prompt under
maxPromptChars.
- 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:
- Check that the output schema is satisfied.
- Check that the output is a single string only.
- Check that
reference_images URLs and referenceNotes are considered when present.
- Check that preservation wording appears when required.
- Check that every input field has a default.
