CO-STAR
Context, Objective, Style, Tone, Audience, Response — the workhorse formula for production prompts.
Use when: Any LLM task where business context matters; default for the Master Engineer.
The 16 prompting formulas
A formula is a contract with the model: it forces the context, the constraint and the output shape into the prompt instead of hoping the model infers them. Each page below gives the skeleton and the situation it is built for.
RTF
Role, Task, Format — minimal but powerful; ideal for tool/function prompts.
Use when: Quick, bounded requests; function-calling prompts.
CRISPE
Capacity & Role, Insight, Statement, Personality, Experiment — creative and strategy prompts.
Use when: Creative writing, strategy, brainstorming, marketing.
TAG
Task, Action, Goal — short and goal-anchored.
Use when: Crisp one-shot tasks; agent instructions.
Few-shot
Provide 2-5 exemplars of the exact input/output shape before asking.
Use when: Format-sensitive output, classification, style mimicry.
Chain-of-Thought
Instruct the model to reason step by step before answering.
Use when: Math, logic, multi-step reasoning; improves accuracy on hard tasks.
Zero-shot CoT
The 'Let's think step by step' trigger — cheapest reasoning boost.
Use when: Any hard question when no examples are available.
Self-Consistency
Sample multiple reasoning paths and take the majority answer.
Use when: High-stakes reasoning; reduces variance.
ReAct
Reason -> Act -> Observe loop; the foundation of agent prompts.
Use when: Agentic tasks with tools; research workflows.
Tree of Thoughts
Explore multiple branches of reasoning and evaluate each.
Use when: Open-ended problem solving, planning.
Persona
Assign an expert persona to raise output quality and consistency.
Use when: Domain-expert output; tone control.
Output Format Contract
Pin the exact response shape: delimiters, JSON schema, length caps.
Use when: Programmatic output, extraction, structured data.
Chain-of-Verification
Generate, then verify each claim, then revise.
Use when: Fact-sensitive answers; hallucination reduction.
Step-Back
Ask for the high-level principle first, then apply it.
Use when: Hard problems where models rush to details.
Positive Instruction + Negative Boundary
State what to do, then what NOT to do — the prompt-side twin of SD negative prompts.
Use when: Safety, style control, compliance.
System Prompt Architect
System: role + constraints + knowledge + process; User: task + context.
Use when: Any fine-tuned system prompt; the Master Engineer's signature output.
The system-prompt architecture
Every formula plugs into the same six-part system-prompt skeleton — this is the structure behind the highest-value paid pack.
| Layer | What it holds |
|---|---|
identity | Who the model is, credentials, and domain. |
constraints | Hard rules: tone, length, safety, format. |
knowledge | What knowledge to rely on; what is out of scope. |
process | The step-by-step method to follow for each task. |
output_contract | Exact response shape: format, schema, delimiters. |
examples | Few-shot exemplars inline when the format is unusual. |