AgentSkillsCN

promptheus

借助 Promptheus MCP 服务器，利用 AI 助力优化和改进提示词。每当用户提出优化、改进、重写或提升提示词的需求时，均可使用此技能。触发短语包括“优化这个提示词”、“改进这个提示词”、“让这个提示词更出色”、“优化这个提示词”、“重写这个提示词”，或任何旨在对现有提示词或指令进行深度打磨、提升其表达效果的请求。此外，当用户仅提供粗略的想法、需求或规格，并希望据此生成一份精心撰写的优质提示词时，此技能同样适用。

SKILL.md
--- frontmatter
name: promptheus
description: AI-powered prompt refinement and improvement using the Promptheus MCP server. Use this skill whenever the user asks to refine, improve, rewrite, or optimize a prompt. Trigger phrases include "refine this prompt", "improve this prompt", "make this prompt better", "optimize this prompt", "rewrite this prompt", or any request that involves taking an existing prompt/instruction and making it more effective. Also triggers when the user provides rough ideas, requirements, or specifications and asks for a well-crafted prompt to be generated from them.

Promptheus Prompt Refinement

Route all prompt refinement requests through the Promptheus MCP tools. Never attempt to refine prompts using your own capabilities -- always delegate to Promptheus.

Workflow

Refine a prompt

  1. •Call mcp__promptheus__refine_prompt with the user's prompt text.
  2. •If the response type is clarification_needed:
    • •Use AskUserQuestion with the questions from questions_for_ask_user_question.
    • •Map user answers to question IDs (q0, q1, q2, ...).
    • •Call mcp__promptheus__refine_prompt again with the original prompt and the answers dict.
  3. •If the response type is refined:
    • •Present the refined prompt to the user.
    • •If the user originally asked to "execute" or "run" the prompt, use the refined prompt to generate the requested content.

Tweak an existing prompt

When the user wants a targeted adjustment to an already-refined prompt (e.g., "make it shorter", "more technical", "add constraints"):

  1. •Call mcp__promptheus__tweak_prompt with the current prompt and the requested modification.
  2. •Present the tweaked result to the user.

Rules

  • •Never refine or rewrite prompts yourself. Always use Promptheus tools.
  • •If Promptheus returns an error, report it to the user and offer to retry or troubleshoot with mcp__promptheus__validate_environment.