Prompts

Colour: Amber (#f59e0b)

Prompts are structured instructions sent to AI models. They are the core building block of most workflows — defining what the model should do, how it should behave, and what output to produce.

Variables

Prompts support variables using the {{VARIABLE_NAME}} syntax. Place them anywhere in your prompt content and they’ll be detected automatically.

Syntax Rules

Format: {{NAME}} — double curly braces around the variable name
Names are alphanumeric plus underscores only (e.g. {{USER_INPUT}}, {{CONTEXT_3}})
Case-sensitive: {{Name}} and {{NAME}} are treated as different variables
Variables are purely user-defined — there are no predefined or system variables

How Variables Are Used

Analysis tab: lists all detected variables with their occurrence count (e.g. {{USER_INPUT}} ×2 if it appears twice)
Test tab: generates an input field for each variable (up to 6). Fill in values and run the test — skrptiq substitutes your values before sending to the LLM
Unfilled variables: if you leave a variable empty when testing, the literal {{VARIABLE}} text passes through to the model
Test case persistence: variable values are saved with each test case in the node’s metadata, so they survive closing and reopening the editor

Example

You are a {{ROLE}} assistant. The user's name is {{USER_NAME}}.

Analyse the following text and provide a {{OUTPUT_FORMAT}} summary:

{{INPUT_TEXT}}

This prompt has 4 variables. In the Test tab, you’d see input fields for ROLE, USER_NAME, OUTPUT_FORMAT, and INPUT_TEXT.

Prompt Templates

When creating a new prompt node, you can start from one of six built-in templates:

Template	Description
System Prompt	Role definition with constraints and output format. Sets up the model’s persona, boundaries, and response structure.
Few-Shot Prompt	System message paired with example input/output pairs. Teaches the model the expected format through demonstration.
Chain-of-Thought	Step-by-step reasoning before the final answer. Forces the model to show its working for complex problems.
Extraction Prompt	Pulls structured data from unstructured input. Returns JSON with typed fields and a confidence score.
Classification Prompt	Categorises input into predefined classes. Returns the chosen category, confidence, and reasoning.
Conversational Prompt	Persona-based conversational agent with tone, knowledge boundaries, and escalation rules.

Each template comes pre-filled with sensible variable placeholders. Pick the closest match and adapt it to your use case, or start from a blank prompt if none fit.

AI Features

Prompts have access to four AI-powered tools in the editor’s right panel. See AI Features for full details.

Analysis — Scores your prompt 0-100 against best practices. Shows a checklist of passing and failing criteria, plus token count and variable statistics.
Test — Create test cases, fill in variables with sample values, run the prompt against an LLM, and auto-generate test inputs.
Review — Expert AI review of your prompt’s quality, clarity, and effectiveness.
Refine — AI-suggested improvements you can accept or reject.

Content Type Detection

The editor detects content type automatically — markdown, code blocks, JSON schemas, and plain text are all handled. This affects syntax highlighting and how the content is displayed.

Connections

Prompts connect to other nodes in the graph:

Used by workflows and skills (via uses edges) — a workflow or skill references the prompt as part of its execution
References sources (via references edges) — a prompt can link to source material it draws on
Derived from other prompts (via derived_from edges) — track prompt lineage when you create variations