返回 Skill 列表
extension
分类: AI Agent 能力无需 API Key

prompt-section-design

设计可组合的提示部分,用于构建具有主动性的提示。在创建可重用的提示组件、设计乐高积木式的提示部分或为利益相关者三部曲构建提示结构时使用。

person作者: jakexiaohubgithub

Prompt Section Design Skill

Design composable prompt sections that work like LEGO blocks for building prompts at any level.

Purpose

Create well-structured prompt sections that are reusable, consistent, and effective for the stakeholder trifecta.

When to Use

  • Creating a new prompt
  • Restructuring existing prompt
  • Adding sections to prompt
  • Standardizing team prompts

Section Tier List

| Tier | Sections | Priority | | --- | --- | --- | | S | Workflow | Always include (Level 2+) | | A | Variables, Examples, Control Flow, Delegation, Template | High value | | B | Purpose, High-Level, Higher Order, Instructions | Supporting | | C | Metadata, Codebase Structure, Relevant Files, Report | As needed |

Design Process

Step 1: Identify Prompt Purpose

Ask:

  • What does this prompt accomplish?
  • Who will use it? (you, team, agents)
  • What level is it? (1-7)
  • What inputs/outputs are needed?

Step 2: Select Required Sections

Based on level:

| Level | Required | Recommended | | --- | --- | --- | | 1 | Title, Prompt | - | | 2 | Title, Workflow | Variables, Report | | 3 | Title, Workflow | Variables, Control Flow | | 4 | Title, Workflow | Variables, Delegation | | 5 | Title, Workflow | Variables | | 6 | Title, Workflow, Template | Variables | | 7 | Title, Workflow, Expertise | Variables |

Step 3: Design Each Section

Metadata (Frontmatter)

---
description: Clear, searchable description
argument-hint: [arg1] [arg2]
allowed-tools: Read, Write, Edit
model: sonnet
---

Guidelines:

  • description: What does it do? When to use?
  • argument-hint: What parameters expected?
  • allowed-tools: Minimal set needed
  • model: Match to task complexity

Title

# Action-Oriented Title

Guidelines:

  • Use imperative verb: Create, Build, Generate, Analyze
  • Be specific: "Create Implementation Plan" not "Plan"
  • Keep concise: 2-5 words

Purpose

## Purpose
[1-2 sentences describing what the prompt accomplishes]

Guidelines:

  • Direct language to agent
  • Reference key sections
  • Explain the "what" and "why"

Variables

## Variables

# Dynamic (from user)
USER_PROMPT: $ARGUMENTS
FILE_PATH: $1
COUNT: $2 or 3 if not provided

# Static (fixed)
OUTPUT_DIR: specs/
MODEL: sonnet

Guidelines:

  • SCREAMING_SNAKE_CASE
  • Dynamic first, static second
  • Include defaults where appropriate
  • Clear descriptions

Workflow (S-Tier)

## Workflow

1. Validate inputs
   - Check USER_PROMPT is provided
   - If not, STOP and ask user
2. Process task
   - Sub-step detail
3. Generate output
4. Report results

Guidelines:

  • Numbered steps for sequence
  • Sub-bullets for details
  • STOP conditions explicit
  • Clear progression

Instructions

## Instructions

- IMPORTANT: Always validate before processing
- Handle edge cases gracefully
- Never modify files outside project

Guidelines:

  • Bullet points for rules
  • IMPORTANT markers for critical
  • Edge cases explicit

Report

## Report

## Task Complete

**Files:** [count]
**Status:** [status]

### Changes
- [change 1]
- [change 2]

Guidelines:

  • Template for output
  • Consistent format
  • Easy to parse

Template (Level 6)

<!-- markdownlint-disable MD033 MD025 MD003 MD040 -->
## Specified Format

```text
---

allowed-tools: <tools>
description: <description>
---

# <name>

## Variables

<VAR>: $1

## Workflow

<steps>
<!-- markdownlint-enable MD033 MD025 MD003 MD040 -->

Guidelines:

  • Complete template
  • Placeholders marked clearly
  • Follows prompt conventions

Expertise (Level 7)

## Expertise

### Domain Knowledge
- Pattern 1 learned
- Pattern 2 discovered

### Discovered Patterns
- Implementation insight 1
- Best practice 2

Guidelines:

  • Organized by category
  • Grows over time
  • Never modify Workflow

Step 4: Validate Structure

Checklist:

  • [ ] Title is action-oriented
  • [ ] Workflow has numbered steps
  • [ ] Variables use SCREAMING_SNAKE_CASE
  • [ ] STOP conditions are explicit
  • [ ] Frontmatter has description
  • [ ] Sections in logical order

Section Order Convention

<!-- markdownlint-disable MD033 MD025 MD040 -->
---

[frontmatter]
---

# [Title]

## Purpose

[purpose]

## Variables

[variables]

## Instructions

[instructions]

## Workflow

[workflow]

## Report

[report format]
<!-- markdownlint-enable MD033 MD025 MD040 -->

Output Format

When designing sections:

## Section Design

**Prompt:** [name]
**Level:** [1-7]

### Recommended Sections

1. **Title**: [suggested title]

2. **Frontmatter**:

```yaml
description: ...
argument-hint: ...
allowed-tools: ...
model: ...
```

1. **Variables**:
   - Dynamic: [list]
   - Static: [list]

2. **Workflow**: [step count] steps
   - Step 1: [overview]
   - Step 2: [overview]
   ...

3. **Report**: [format type]

Anti-Patterns

| Anti-Pattern | Problem | Solution | | --- | --- | --- | | No Workflow section | Agent lacks direction | Always add for Level 2+ | | Inconsistent variable names | Confusion | SCREAMING_SNAKE_CASE | | Missing STOP conditions | Runaway execution | Explicit early exits | | Over-detailed workflow | Reduces agent autonomy | High-level steps | | No frontmatter | Hard to discover | Add description |

Key Quote

"Build libraries of reusable battle-tested agentic prompts with composable sections that work like LEGO blocks."

Cross-References

  • @prompt-sections-reference.md - Section definitions
  • @seven-levels.md - Sections by level
  • @variable-patterns.md - Variable conventions

Version History

  • v1.0.0 (2025-12-26): Initial release

Last Updated

Date: 2025-12-26 Model: claude-opus-4-5-20251101