Skills

Colour: Indigo (#6366f1)

Skills are reusable units of functionality or logic — the “doing” nodes. Each skill represents a discrete capability: summarisation, translation, classification, data transformation, entity extraction, and so on.

When to use skills

Use a skill node when you want to capture a repeatable capability that could be used across multiple workflows or contexts. If you find yourself describing what something does rather than what it says, it’s probably a skill.

Skills vs prompts: A skill describes the capability (e.g. “Summarise Meeting Notes”). A prompt contains the exact template text sent to an LLM to execute that capability. Skills often connect to one or more prompts via “uses” connections.

Skills vs workflows: A skill is a single unit of work. A workflow orchestrates multiple skills (and prompts, services, sources) into a sequence. Think of skills as individual steps and workflows as the full pipeline.

How skills connect

Skills typically appear in the middle of a graph’s dependency chain:

• Workflows → Skills — workflows use skills as steps in a sequence (connection type: “uses”)
• Skills → Prompts — skills use prompts to define the LLM instructions (connection type: “uses”)
• Skills → Skills — one skill can depend on another (connection type: “depends_on”)
• Skills → Sources — skills can reference guidelines or standards (connection type: “references”)
• Skills → Services — skills can require specific LLM providers (connection type: “runs_on”)

Examples

• Summarise Meeting Notes — takes raw meeting transcripts and produces structured summaries
• Extract Action Items — identifies tasks and owners from unstructured text
• Classify Intent — categorises user messages by intent for routing
• Translate to French — language translation with tone preservation
• Generate SQL from Natural Language — converts plain English queries to SQL

Editor

Skills use the standard node editor — title, description, content, tags, and type selector. There are no skill-specific editor features. The content field is free-form: use it for implementation notes, pseudocode, capability descriptions, or anything that helps document what the skill does.