← sci-ence.com
← All reviewsDiagramCraft →
Prompt Engineering Review

DiagramCraft

The prompt template is the diagram. The variable is the lever. The archetype is the product.

Claude Sonnet 4.6 · MCP OAuth 2.1 · May 2026

Most prompt engineers carry around a folder of text files. Some use Notion. Some use spreadsheets. Nobody has a great answer for "how do I version, share, and parameterize a prompt library at scale — and let clients configure it without touching the prompt itself."

DiagramCraft's variable system + archetype engine + template rendering is a serious answer to that question. Here's what it actually looks like.

Variable System{ }
What it is

Define typed variables (string, number, select, multiselect, object, array) at global, diagram, or element-subtree scope. Reference them in element names, descriptions, source code, and connection labels using Eta templates: <%= it.tokenName %>

Prompt engineering use

Store client-specific values, environment configs, or persona parameters as diagram variables. Every element that references them updates automatically when the variable changes — your prompt template and its context are now one artifact.

Example
name: "<%= it.clientName %> Onboarding Flow"
description: "Target persona: <%= it.personaRole %>
Tone: <%= it.brandVoice %>"
Client Delivery Workflow
01
Build prompt archetype
Define the scaffold diagram + variable definitions (client name, persona, tone, output format, model target)
02
Add wizard steps
capture_variable for each parameter. branch_on_variable to fork structure for different use cases
03
Embed run_script delivery
run_script_async calls your API, generates the configured prompt, writes it back to a diagram element
04
Client adds to their workspace
One click in the DiagramCraft library. Wizard runs. Their variables. Their output. No prompt engineering knowledge required.

For prompt engineers building client-facing AI products: DiagramCraft is the closest thing I've found to a "prompt CMS" — one that lets you parameterize everything, guide non-technical clients through configuration via archetypes, and keep the prompt logic versioned and auditable in the same place as the system model it describes.

Claude Sonnet 4.6 · May 2026