# Prompt Creation

> Use when writing a new system prompt from scratch, structuring it with Anthropic's 9-element format, or choosing emphasis levels and few-shot examples.

- Skill: `fusengine/prompt-creation` (Agent Skill, multi-file: 5 files)
- Install (CLI): `npx skillmds@latest add fusengine/prompt-creation`
- Raw SKILL.md: https://api.skillmd.com/api/skills/fusengine/prompt-creation/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: fusengine (https://skillmd.com/u/fusengine)
- Updated: 2026-09-10
- Page: https://skillmd.com/skills/fusengine/prompt-creation

---


<objective>
Prompt Creation covers Anthropic's official 9-element system-prompt structure (task context, tone, task description + rules, few-shot examples, input data, immediate task, precognition/scratchpad, output formatting, prefill) and the 6-level emphasis scale from "Please do X" up to "CRITICAL - ZERO TOLERANCE". It also covers the 8-step creation workflow from defining the objective through iterating after mental edge-case testing.

For RAG/retrieval-augmented prompts specifically, see `rag-patterns.md`; for prompts combining images with text (vision, screenshot-to-code, diagram understanding), see `multimodal-prompting.md`.
</objective>

# Prompt Creation

Skill for creating high-performance prompts following 2025 best practices and Anthropic conventions.

## Documentation

- [techniques.md](docs/techniques.md) - Structuring techniques (CoT, Few-shot, etc.)
- [templates.md](docs/templates.md) - Reusable templates
- [rag-patterns.md](references/rag-patterns.md) - Load when: designing a RAG/retrieval-augmented prompt (chunking strategy, context assembly, citation injection, hallucination prevention)
- [multimodal-prompting.md](references/multimodal-prompting.md) - Load when: writing a prompt that combines images/screenshots/diagrams with text (vision+text, screenshot-to-code, diagram understanding)

## Anthropic Official Structure (9 Elements)

Each system prompt should cover these 9 aspects:

### 1. Task Context
```markdown
# Task Context
You are an expert [ROLE] specializing in [DOMAIN].
Your goal is to [PRIMARY_OBJECTIVE].
```

### 2. Tone Context
```markdown
# Tone
- Tone: [formal/casual/technical/friendly]
- Length: [concise/detailed/adaptive]
- Format: [prose/lists/tables/mixed]
```

### 3. Task Description + Rules
```markdown
# Task Description
Here are important rules:
- Always [MANDATORY_BEHAVIOR]
- Never [FORBIDDEN_BEHAVIOR]
- If [CONDITION] then [ACTION]

If unsure, say "I don't have enough information to answer."
```

### 4. Examples (Few-shot)
```markdown
# Examples

<example>
Input: [STANDARD_CASE_INPUT]
Output: [EXPECTED_OUTPUT]
</example>

<example>
Input: [EDGE_CASE_INPUT]
Output: [EXPECTED_OUTPUT]
</example>
```

### 5. Input Data
```markdown
# Input Data
<document>
[USER_PROVIDED_CONTENT]
</document>
```

### 6. Immediate Task
```markdown
# Immediate Task
Now, [SPECIFIC_ACTION] the above [document/code/data].
```

### 7. Precognition (Scratchpad)
```markdown
# Precognition
First, in <scratchpad> tags, analyze the key points.
Then provide your final answer in <answer> tags.
```

### 8. Output Formatting
```markdown
# Output Format

## Analysis
[Problem understanding]

## Solution
[Main response]

## Recommendations
[Additional suggestions]
```

### 9. Prefill (Assistant Turn)
```markdown
# Prefill
<scratchpad>
```

## Emphasis Techniques

In order of increasing effectiveness:

| Level | Syntax | Usage |
|-------|--------|-------|
| 1 | `Please do X` | Suggestion |
| 2 | `You should do X` | Recommendation |
| 3 | `Always do X` | Standard rule |
| 4 | `YOU MUST do X` | Strong rule |
| 5 | `IMPORTANT: Do X` | Critical rule |
| 6 | `CRITICAL - ZERO TOLERANCE: Do X` | Absolute rule |

## Creation Workflow

```
1. DEFINE the objective (1 sentence)
   ↓
2. IDENTIFY target audience
   ↓
3. LIST constraints
   ↓
4. CHOOSE techniques (CoT? Few-shot?)
   ↓
5. WRITE with 9-element structure
   ↓
6. ADD guardrails
   ↓
7. TEST mentally (edge cases)
   ↓
8. ITERATE if necessary
```

## Forbidden

- Never create prompts without clear objective
- Never write ambiguous instructions
- Never forget error cases
- Never ignore target model context