AgentSkillsCN

fabric-improve-prompt

运用提示工程的最佳实践，优化LLM提示。适用于用户希望优化提示、改进Slash命令，或应用提示工程技巧时使用。

SKILL.md
--- frontmatter
name: fabric-improve-prompt
description: Improve LLM prompts using prompt engineering best practices. Invoke when user wants to optimize prompts, improve slash commands, or apply prompt engineering.
allowed-tools: ["Read", "Write", "Edit", "Bash"]

Improve Prompt

Apply prompt engineering best practices to improve an LLM prompt for better, more consistent results.

Purpose

Take an existing prompt (from a slash command, CLAUDE.md, skill, or ad-hoc use) and improve it using proven prompt engineering strategies:

  • •Add specificity and context
  • •Improve structure and formatting
  • •Add examples (few-shot)
  • •Clarify output expectations
  • •Set appropriate persona/role
  • •Add constraints and guardrails

Input Sources

User provides one of:

  • •Pasted prompt: Directly in message
  • •File reference: Path to a slash command or pattern file
  • •Command name: Name of existing slash command to improve

Process

1. Get the Prompt

From pasted text: Use the text directly.

From file: Read the file using the Read tool.

From slash command: Read from the user's slash-commands directory.

2. Analyze and Improve

Apply prompt engineering principles:

  1. •Be Specific: Add relevant details and constraints
  2. •Clear Structure: Use numbered steps and delimiters
  3. •Provide Examples: Add few-shot examples when helpful
  4. •Specify Output: Define expected format explicitly
  5. •Set Persona: Define expertise when specialized knowledge helps
  6. •Add Constraints: Length limits, what to avoid
  7. •Request Reasoning: "Think step by step" for complex tasks
  8. •Reference Material: Include sources when accuracy matters
  9. •Decompose Tasks: Break complex into subtasks
  10. •Edge Cases: Consider boundary conditions

3. Output Format

markdown
## Analysis

- [Issue 1]: [Description]
- [Issue 2]: [Description]
- [Issue 3]: [Description]

## Improved Prompt

[Ready-to-use improved prompt]

code

## Changes Made

- [Change]: [Why it helps]
- [Change]: [Why it helps]

## Usage Notes

[Tips for using effectively]

4. Optional: Apply Changes

If improving a slash command or pattern file, offer to apply changes:

"Would you like me to update the file with this improved prompt?"

If yes, use Edit tool to update the file while preserving YAML frontmatter and structure.

Usage Examples

Improve pasted prompt:

code
/fabric-improve-prompt

You are a code reviewer. Review this code and tell me if it's good.

Improve existing slash command:

code
/fabric-improve-prompt /my-command

Improve from file:

code
/fabric-improve-prompt ~/my-patterns/summarize.md

When NOT to Over-Engineer

  • •Simple, one-off questions don't need elaborate prompts
  • •If the original prompt is already good, say so
  • •Don't add complexity that doesn't improve results
  • •Preserve original intent - don't change the goal

Error Handling

No prompt provided:

  • •Ask user to paste a prompt or specify a file/command

File not found:

  • •Report error, suggest checking path

Already optimized:

  • •Note that prompt is already well-structured
  • •Suggest minor refinements if any

Philosophy: Good prompts are the leverage point for AI effectiveness. A 10-minute investment in prompt improvement can save hours of poor results.