CO-STAR

What It Is

Context, Objective, Style, Tone, Audience, Response. A structural template popularised by Sheila Teo, winner of Singapore GovTech's 2023 GPT-4 prompt engineering competition.

When to Use It

Use for any content or analysis request where audience and format matter. Skip it for pure reasoning or tool-use tasks.

Blank Template

# CONTEXT
{{background}}
# OBJECTIVE
{{what you want done}}
# STYLE
{{writing style or expert persona}}
# TONE
{{tone}}
# AUDIENCE
{{who reads it}}
# RESPONSE
{{format, length, structure}}

Five Sample Prompts for Gen AI Professionals

COSTAR-01 - Model selection brief

# CONTEXT
We run a customer-support assistant handling {{volume}} tickets a month. Current model: {{current_model}}. Latency budget {{latency_ms}} ms, cost ceiling {{monthly_budget}}.
# OBJECTIVE
Recommend whether to switch models and how to prove it.
# STYLE
Senior ML engineer writing a decision memo.
# TONE
Direct, evidence-first.
# AUDIENCE
Engineering manager and finance lead.
# RESPONSE
Comparison table of 3 candidate models, stated assumptions, one recommendation, a 2-week test plan. Under 600 words.

COSTAR-02 - Eval report summary

# CONTEXT
<eval_results>{{paste results}}</eval_results>
# OBJECTIVE
Summarise what changed between model versions {{v_old}} and {{v_new}}.
# STYLE
Release-notes style.
# TONE
Neutral, no hype.
# AUDIENCE
Product managers with no ML background.
# RESPONSE
5 bullets of wins, 5 bullets of regressions, one go/no-go line.

COSTAR-03 - RAG architecture explainer

# CONTEXT
Our internal knowledge base has {{doc_count}} documents in {{formats}}. Users ask {{query_types}}.
# OBJECTIVE
Explain the proposed RAG architecture and its trade-offs.
# STYLE
Architecture decision record.
# TONE
Plain and precise.
# AUDIENCE
Back-end team who will build it.
# RESPONSE
Diagram described in text, component list, 3 alternatives rejected with reasons.

COSTAR-04 - Client AI pilot proposal

# CONTEXT
Client: {{industry}} firm, {{headcount}} staff, pain point: {{pain_point}}.
# OBJECTIVE
Draft a 6-week AI pilot proposal.
# STYLE
Management-consulting proposal.
# TONE
Confident, realistic about risks.
# AUDIENCE
COO and IT head.
# RESPONSE
Problem, approach, weekly plan, success metrics, risks, price range placeholder. 1 page.

COSTAR-05 - Local LLM onboarding guide

# CONTEXT
Team laptops: {{ram_gb}} GB RAM, {{gpu}}. Data cannot leave the device.
# OBJECTIVE
Write an onboarding guide to run a local LLM for coding help.
# STYLE
Step-by-step technical guide.
# TONE
Friendly, no jargon without a definition.
# AUDIENCE
Developers new to local AI.
# RESPONSE
Numbered steps, a model-size table by RAM, troubleshooting section.

Links

• Up: Frameworks
• Curated by: curation prompts