Token导航 LogoToken导航TokenDH.com
开发只读clawhub未标认证来源可访问clear审计通过

prompt-optimizer-en提示优化器 en

Agent Skill

用于辅助提示词、系统指令、Agent 行为约束和工作流模板的整理。它适合让 Agent 规范任务边界、统一输出格式、拆分操作步骤或优化提示词可复用性。使用时需要保留真实业务约束,不要把示例当硬规则;涉及自动执行、外部工具或高风险操作时,应在提示词中明确确认步骤、权限边界和失败处理方式。

总安装

3,672

周安装

150

GitHub Stars

公开资料未说明

下载量

1,188
OpenClaw

安装说明

本站只整理中文说明和来源信息,不托管安装包,也不代用户安装。

GitHub

来源数

2

许可证

MIT-0

最后核验

2026-05-01

来源状态

来源可访问

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:prompt-optimizer-en(提示优化器 en)
来源仓库:https://github.com/ucsdzehualiu/prompt-optimizer-en
安装命令:
openclaw skills install prompt-optimizer-en
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 OpenClaw 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

ClawHubOpenClaw
openclaw skills install prompt-optimizer-en

简介

两阶段迭代优化与自动提示工程的复合优化器。

  • 适用于复杂任务的 ACON 框架严格执行场景。
  • 结合 APE 技术实现提示质量的持续提升。
  • 需设定明确的优化目标与终止条件以控制迭代过程。
  • 安装方式:通过 clawhub 平台使用 openclaw skills install prompt-optimizer-en 命令部署。

SKILL.md

name
prompt-optimizer
description
Iterative prompt optimizer for complex tasks. Strictly implements ACON's two-stage iterative optimization + APE automatic prompt engineering. Only triggers when user explicitly requests it, actively collects feedback after optimization, supports multi-round iteration until satisfied.
usage
Only activate when user explicitly says "optimize prompt", "improve prompt", "refine instruction", never auto-trigger.
author
Based on arXiv:2510.00615 (ACON), arXiv:2211.01910 (APE)
license
MIT
tags

Atomic Optimization Methodology

🔬 Stage 1: Input Parsing & Critical Signal Extraction (ACON Paper §3.1)

Input: User's original prompt Operations:

  1. Intent Locking: Extract core task goal T, ensure all subsequent optimizations never deviate from T
  2. Critical Signal Extraction (ACON-defined mandatory signals):

- ✅ Role Definition R: Expert role specified by user - ✅ Task Goal T: What the core task is - ✅ Constraints C: Boundary rules, prohibitions - ✅ Output Format F: Output structure/format requested by user - ✅ Variable Placeholders V: All {{variable_name}} - ✅ Examples E: Few-shot examples provided by user - ✅ Tool Rules U: When and how to use tools - ✅ Success Criteria S: What constitutes a good output

  1. Baseline Measurement: Record original prompt token length L₀

🚀 Stage 2: APE Utility Enhancement (arXiv:2211.01910 Automatic Prompt Engineering)

Goal: Turn vague prompts into expert-level instructions, improve utility Operations (Strict Order):

  1. Candidate Generation: Based on original prompt, generate 5 candidate instructions in different styles

- Candidate 1: Structured instruction version - Candidate 2: Expert role version - Candidate 3: Constraint reinforcement version - Candidate 4: Format clarification version - Candidate 5: Logic optimization version

  1. Candidate Scoring (APE paper scoring mechanism):

- Clarity: Are instructions clear and unambiguous (0-10) - Completeness: Does it include all critical signals (0-10) - Effectiveness: Can it guide the model to produce high-quality output (0-10)

  1. Optimal Selection: Choose the candidate with highest total score, as utility-enhanced version P₁
  2. Validation: Verify P₁ 100% preserves all critical signals, no change to original intent

📦 Stage 3: ACON Compression Optimization (ACON Paper §3.3 Two-Stage Optimization)

Goal: Compress token length without breaking functionality Operations (Strict Order: Utility first, then compression):

  1. Redundancy Analysis: Analyze redundant content in P₁

- Duplicate instructions and requirements - Fluff, jargon, ineffective expressions - Verbose statements that can be simplified

  1. Selective Compression:

- Only remove redundancy, NEVER delete critical signals - Merge duplicate content - Rewrite with more concise language, keep semantics unchanged

  1. Functional Equivalence Validation:

- Ensure compressed P₂ is functionally identical to P₁ - Ensure all critical signals are fully preserved - Ensure no change to original task goal

  1. Length Control: Adjust compression degree based on λ parameter (performance-cost tradeoff)

- Default λ=0.5: Balanced mode - If user feedback "too long", automatically increase λ to 0.8 for more compression - If user feedback "not effective enough", automatically decrease λ to 0.2 to reduce compression


📤 Stage 4: Output & Feedback Collection

Operations:

  1. Output optimized prompt P₂, wrapped in code block for easy copying
  2. Actively ask for user feedback:
   Optimization complete. Does this version meet your needs?
   If there's anything unsatisfactory, please let me know, such as:
   - Not effective enough?
   - Still too long?
   - Some constraints/formats not preserved?
   - Other issues?
   I'll continue iterating based on your feedback.

🔄 Stage 5: Iterative Optimization (ACON Paper's R-round Iteration Mechanism)

When user provides feedback, execute the following:

  1. Feedback Parsing: Identify feedback type

- Type A: Not effective enough → Go back to Stage 2, re-run APE utility enhancement, add constraints - Type B: Too long → Go back to Stage 3, re-run ACON compression, increase λ - Type C: Some content not preserved → Check critical signals, restore missing parts - Type D: Other requirements → Adjust based on user's specific request

  1. Re-run Optimization: Adjust parameters based on feedback, run two-stage optimization again
  2. Validation: Ensure new version preserves core task goal, and solves the user's feedback issue
  3. Output new optimized version, ask for feedback again
  4. Repeat until user indicates satisfaction

Strict Rules (Guarantee Effectiveness)

  • ✅ Every step has validation, ensure no damage to original functionality
  • ✅ Critical signals are NEVER deleted, 100% preserved
  • ✅ Strictly follow "utility first, then compression" order, never reverse
  • ✅ Each iteration re-validates, ensure it gets better with each round
  • ✅ For complex tasks, prioritize functional integrity, compression is optional
  • ❌ Never auto-trigger, only work when user explicitly requests
  • ❌ No comparisons or analysis, only output optimized results
  • ❌ No extra explanations unless explicitly requested

适合场景

01

OpenClaw 用户查找和安装 Skill 时

02

用户想查找某类 Agent Skill 时

03

需要根据任务场景推荐可安装能力包时

04

需要对比不同来源的安装命令和来源信息时

能力概览

能力 1

按任务关键词查找相关 Skills

能力 2

展示可复制的安装命令

能力 3

保留来源站点、仓库和原始说明,方便继续核验

能力 4

补充不同宿主或平台的使用分布数据

能力 5

展示第三方安全扫描或审计结果

安装后应在对应宿主中按原始 README 的触发条件使用;具体调用方式请以来源页面和 README 为准。

平台分布

OpenClaw

71.47%
按下载量换算849

安全审计

VirusTotal

通过

ClawScan

通过

Static analysis

通过

权限和风险

只读

该 Skill 主要提供规则、说明或参考内容,本身偏只读;真正读写文件、联网或执行命令仍取决于宿主 Agent 的任务。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。当前只有一个来源,正式发布前建议补源仓库或其他目录站核验。

来源信息

继续浏览同类 Skills