/craft — Structured prompt generation

A prompt that creates a prompt. Give it a topic; it produces a fully structured prompt using the C.R.A.F.T. framework that you then paste into a fresh LLM conversation to get exceptional output.

/craft is also a built-in step in /linear:plan-work --craft — it sharpens your issue description into a clearer prompt before the agent drafts the issue. Without it, research mode works from whatever you typed; with it, the agent works from a refined, structured description.

The CRAFT framework

Each letter represents a section of the generated prompt:

Section	Purpose	Key requirement
C — Context	Frame the situation and what knowledge the LLM should reference	Comprehensive: goals, expertise areas, domain knowledge, references
R — Role	Define the LLM's persona, experience, and expertise level	Industry-leading expert with 20+ years of relevant experience
A — Action	Numbered sequential steps the LLM should follow	Ordered list that maximises success; include "fill in the blank" elements where the user hasn't provided specifics
F — Format	Structural arrangement of the output	Specify: essay, table, code, markdown, summary, list, etc.
T — Target audience	The ultimate consumer of the output	Demographics, geography, language, reading level, preferences

How it works

1. You give /craft a topic — "create a guide to setting monthly goals"
2. The skill drafts all five CRAFT sections for that topic
3. Placeholders ([bracketed]) are added where you haven't specified details
4. You get a complete prompt to paste into a fresh LLM conversation

The two-step workflow is deliberate — the CRAFT prompt generates the actual output in a clean context, free from conversation history.

When to use

• Creating a prompt for any LLM (ChatGPT, Claude, Gemini, etc.)
• Turning a vague idea into a comprehensive, structured prompt
• /linear:plan-work --craft uses it automatically to sharpen issue descriptions before drafting

Key principles

Principle	What it means
Comprehensive detail	The best prompts leave nothing to question — goals, expertise, domain knowledge, format, audience, references, examples
Fill in the blanks	When the user hasn't specified something, add [bracketed placeholders] they can populate
Two-step workflow	Step 1: topic → CRAFT prompt. Step 2: paste prompt into fresh conversation → actual output
Role elevation	Always make the role an industry-leading expert — this pushes the LLM past average responses
Sequential actions	The action section must be a numbered list of steps in the order that maximises success

Example

Topic: "Create a guide to setting and achieving monthly goals"

The skill produces a prompt with:

• Context framing the guide's purpose (breaking objectives into actionable monthly steps, SMART goals)
• Role as a productivity coach with 20+ years of experience
• Action as a 7-step sequence (introduction → breakdown → priorities → tracking → examples → obstacles → conclusion)
• Format specifying plain text with headings, numbered lists, and practical examples
• Target audience defining working professionals aged 25-55, 6th grade reading level

You paste this into a fresh conversation and get output that far exceeds a one-line "write me a guide about monthly goals" prompt.

Source

Based on the C.R.A.F.T. prompt framework — a meta-prompting technique for generating structured, high-quality LLM prompts. See The best ChatGPT Prompt I've ever created for the original source.