ArthurZakirov Architecture Diagram GPT Image Generator
Produce a GPT prompt that, when pasted into a chat model, generates a polished architecture or workflow diagram image. Write the prompt to a file. Keep chat output to the file path only.
Output Discipline
When this skill fires:
- Write the full prompt to
~/diagram-prompts/<slug>.md. Create the directory if missing.<slug>is short kebab-case inferred from the topic. - The file content must be the prompt and nothing else.
- The chat reply must be exactly:
Prompt -> /home/arthur/diagram-prompts/<slug>.md
Open the file, copy all, paste into GPT.
- Do not echo the prompt body back into chat after writing it.
Minimum Input
The user may give:
- A Mermaid diagram
- A rough voice-note style description
- A list of services and connections
- A request to diagram the current project
If critical flow direction or system intent is genuinely ambiguous, ask one short clarifying question. Otherwise fill gaps from the defaults below.
Mandatory Defaults
Always apply these unless the user explicitly overrides them:
- Orientation defaults to horizontal left-to-right.
- Image aspect ratio defaults to 16:9 horizontal landscape.
- The prompt must be self-contained and usable in a fresh chat session.
- Mermaid source inside the prompt should default to
flowchart LR.
Prompt Template
Write the prompt file using this structure, adapting the content to the specific diagram:
Create a polished enterprise architecture diagram image. Render the image in 16:9 horizontal landscape aspect ratio. Do NOT output Mermaid code. Do NOT explain the diagram. Your task is to GENERATE AN IMAGE based on the specification below.
## Layout & system boundary
- Image canvas: 16:9 horizontal landscape
- Horizontal left-to-right layout
- [system-boundary containers and spatial arrangement]
## Visual style
- Clean, modern, enterprise-grade architecture diagram
- Rounded containers, soft shadows, generous spacing
- Crisp directional arrows; label branches on every decision diamond
- Minimal text inside nodes — use icons or emojis for meaning
- Prioritize clarity over technical detail
## Logos & icon mapping
Use recognizable logos for:
- [services that appear in the diagram]
Use visual metaphors:
- [emojis used in the diagram]
## Flow structure
[numbered left-to-right stages, grouping nodes and decisions]
## Color guidance
- System container: [platform tint]
- Entry points or triggers: warm red or amber
- Inputs or read tools: green
- AI or prompt nodes: purple or blue
- Decisions or retry loops: neutral blue-gray
- Output channels: green
- Per-service nodes use brand colors
## Semantic constraints
- [topology, ordering, and decision rules]
## Source structure (for reference only, do not output)
```mermaid
[mermaid source from the user or inferred from the description]
## Default Logo Library
Map products and systems to recognizable logos when relevant:
- Jira
- Bitbucket
- Slack
- Databricks
- AWS
- Kubernetes
- Datadog
- Confluence
- OpenAI
- Claude
- GitHub
- Google Sheets
- Java or backend services
If a service is not covered, use a plain rounded box with a fitting emoji.
## Default Visual Metaphors
Use emojis only when a concept has no natural product logo. Useful defaults include:
- `📄` prompt or document
- `🏷️` classification or label
- `💬` rationale or comment
- `⚙️` decision logic or processing
- `🛠️` fix or modification
- `🔒` validation
- `📣` routing
- `🚨` alert or failure
- `✅` success
- `❌` failure
- `⏰` schedule
- `🔁` retry loop
- `🔍` search or lookup
- `📊` metrics
- `🧠` AI or agent
- `🗄️` storage
- `📥` input
- `📤` output
## Parsing The User's Input
1. Identify services and products, then map them to logos.
2. Identify flow direction. Default to horizontal LR except genuine hierarchy-heavy exceptions.
3. Identify containers or subgraphs.
4. Identify decision diamonds and label both branches.
5. Identify retry loops and label them clearly.
6. Apply colors by role.
7. Build a Mermaid source block so the generator has a spatial spine.
## File Writing
Create `~/diagram-prompts` if it does not exist, then write the full prompt to `~/diagram-prompts/<slug>.md`. Do not add timestamps unless the user explicitly asks.
## Final Chat Reply
After the file is written, the chat reply must be exactly:
```text
Prompt -> /home/arthur/diagram-prompts/<slug>.md
Open the file, copy all, paste into GPT — or reply "imagegen" to render it directly via Codex CLI.
Optional Codex Image Generation
If the user explicitly replies with imagegen, yes generate, render it, or another clear affirmative, run the local helper that turns the prompt into an image and return only the image path plus a one-line open command.