mcpField
P
prompt-creation

prompt-creation

github.com/fusengineAgent workflows3458 installsUpdated 2w ago
Repository

About prompt-creation

A prompt-writing skill based on Anthropic's system-prompt structure. It guides objective setting, prompt composition, emphasis, and iterative edge-case checks.

Key features

  1. 01Nine-part system-prompt structure
  2. 02Six-level instruction emphasis scale
  3. 03Eight-step prompt creation workflow
  4. 04Few-shot and other structuring techniques
  5. 05Separate guidance for RAG and multimodal prompts

Use cases

  1. 01Drafting a structured system prompt
  2. 02Clarifying rules and output requirements
  3. 03Designing prompts for retrieval or image input

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/fusengine/agents --skill prompt-creation

After installing, type/prompt-creationin Claude Code to confirm it loaded.

Bundled files

References and scripts installed along with SKILL.md.

docs/
techniques.md
templates.md
references/
multimodal-prompting.md
rag-patterns.md

Details

From the project README
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.

Prompt Creation

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

Documentation

  • techniques.md - Structuring techniques (CoT, Few-shot, etc.)
  • templates.md - Reusable templates
  • rag-patterns.md - Load when: designing a RAG/retrieval-augmented prompt (chunking strategy, context assembly, citation injection, hallucination prevention)
  • 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
# Task Context
You are an expert [ROLE] specializing in [DOMAIN].
Your goal is to [PRIMARY_OBJECTIVE].
2. Tone Context
# Tone
- Tone: [formal/casual/technical/friendly]
- Length: [concise/detailed/adaptive]
- Format: [prose/lists/tables/mixed]
3. Task Description + Rules
# 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)
# Examples

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

<example>
Input: [EDGE_CASE_INPUT]
Output: [EXPECTED_OUTPUT]
</example>
5. Input Data
# Input Data
<document>
[USER_PROVIDED_CONTENT]
</document>
6. Immediate Task
# Immediate Task
Now, [SPECIFIC_ACTION] the above [document/code/data].
7. Precognition (Scratchpad)
# Precognition
First, in <scratchpad> tags, analyze the key points.
Then provide your final answer in <answer> tags.
8. Output Formatting
# Output Format

## Analysis
[Problem understanding]

## Solution
[Main response]

## Recommendations
[Additional suggestions]
9. Prefill (Assistant Turn)
# Prefill
<scratchpad>

Emphasis Techniques

In order of increasing effectiveness:

LevelSyntaxUsage
1Please do XSuggestion
2You should do XRecommendation
3Always do XStandard rule
4YOU MUST do XStrong rule
5IMPORTANT: Do XCritical rule
6CRITICAL - ZERO TOLERANCE: Do XAbsolute 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

Comments

Markdown supported