01
Framework

CO-STAR

The Singapore GovTech framework that won the world's first GPT-4 prompting competition.

Context · Objective · Style · Tone · Audience · Response Format
Origin

Sheila Teo published CO-STAR in late 2023 after winning the Singapore GPT-4 Prompt Engineering World Championship organized by GovTech. The framework codifies what high-stakes business writing teams already do intuitively — separating context from objective from voice — and forces the model to honor each layer instead of collapsing them.

MicrosoftIBM ConsultingSingapore GovTechFortune-500 enablement programs
The anatomy

Every CO-STAR prompt has these sections.

01

Context

Background facts the model needs before reasoning — domain, constraints, prior decisions, what's already been tried.

02

Objective

The single, measurable outcome. Not 'help with marketing' but 'draft a 200-word LinkedIn post that drives demo signups'.

03

Style

The writing register — McKinsey memo, technical RFC, Hemingway-tight, conversational Slack message.

04

Tone

Emotional posture — confident, empathetic, urgent, neutral-analytical, playful.

05

Audience

Exactly who reads this — CFO, junior dev, skeptical procurement lead, 12-year-old curious about space.

06

Response Format

Hard schema — JSON keys, markdown headings, table columns, exact length, what to omit.

When to use
  • Customer-facing communication where tone is as important as content (sales emails, support replies, exec memos).
  • Marketing and brand voice work where audience and style must stay consistent across many outputs.
  • Anything that needs a strict response format (JSON, fixed-length, specific markdown structure).
  • Translating a vague stakeholder request into a precise model brief.
When not to use
  • ×Pure step-by-step procedural tasks (use RISEN — its 'Steps' slot is built for this).
  • ×Long technical chains running on Claude where XML tags help (use Anthropic XML).
  • ×GPT-5 production prompts that need strict role-priority hierarchy (use OpenAI Developer Message).
Best models
GPT-4o / GPT-5Claude Sonnet 4.5Gemini 2.5 Pro
Worked example

From one sentence to a CO-STAR section.

Input

"Write me a follow-up to a prospect who went silent after a demo."

Sample CO-STAR section
## Context
We demoed our analytics platform 11 days ago to the Director of RevOps at a 400-person SaaS in Series C. They asked good questions, requested pricing, and have not replied to two follow-ups…

Ready to write your first CO-STAR prompt?

Drop a sentence. The Forge analyzes your task, asks the right clarifying questions, then generates a complete enterprise meta prompt structured exactly to CO-STAR.