Prompt Optimizer

Dual-Engine Prompt Optimization Skill — Template matching (3,344 library) + LLM meta-prompting (host AI as optimizer). Zero API cost.

---

🖥️ Platform Compatibility (Read This First!)

Different platforms have different capabilities. Find yours below and follow the recommended setup.

⭐⭐⭐⭐⭐ OpenClaw (Best Experience)

Feature	Status
Template Engine	✅ Full support
LLM Engine	✅
Category Loading	✅
State Persistence	✅ Auto-save config
Quality Evaluation	✅

Install: Place in skills/prompt-optimizer/ directory. OpenClaw auto-detects. Recommended Mode: Template-first + LLM fallback (default). All features available. No limitations.

---

⭐⭐⭐⭐ Claude Code

Feature	Status
Template Engine	✅ Reads local files
LLM Engine	✅
Category Loading	✅
State Persistence	⚠️ In-conversation only; lost on restart
Quality Evaluation	✅

Install: Copy skill folder into your project. Add to CLAUDE.md: Read and follow skills/prompt-optimizer/SKILL.md Recommended Mode: Template-first (default), in-conversation state memory. Limitation: Must re-enable ("enable prompt optimizer") each new conversation.

---

⭐⭐⭐½ Cursor

Feature	Status
Template Engine	✅ Reads local files
LLM Engine	✅
Category Loading	✅
State Persistence	❌ No cross-session persistence
Quality Evaluation	✅

Install: Copy skill folder into project root. Add to .cursorrules: Read and follow skills/prompt-optimizer/SKILL.md. Activate when user says "enable prompt optimizer". Recommended Mode: Template-first (default). Limitation: Must re-enable each session; manual file copy required.

---

⭐⭐⭐ Gemini CLI

Feature	Status
Template Engine	⚠️ Large files (>1MB) may truncate
LLM Engine	✅
Category Loading	⚠️ Small categories recommended
State Persistence	❌ None
Quality Evaluation	✅

Install: Same as Cursor — place in project, reference in system instructions. Recommended Mode: "set engine to LLM-only" or load small categories only (Tech Tools 36KB, Translation 57KB). Limitation: Free tier context is limited; large JSON may truncate. LLM-only mode recommended.

---

⭐⭐⭐½ Cline / AI Coding Assistants

Feature	Status
Template Engine	✅ Reads local files
LLM Engine	✅
Category Loading	✅
State Persistence	❌ No cross-session persistence
Quality Evaluation	✅

Install: Copy skill folder into project. Reference SKILL.md in Cline's custom instructions. Recommended Mode: Template-first (default). Limitation: Must re-enable each session; manual file copy required.

---

⭐⭐ ChatGPT Custom GPT

Feature	Status
Template Engine	❌ Cannot read JSON files
LLM Engine	✅
Category Loading	❌
State Persistence	❌
Quality Evaluation	⚠️ No baseline comparison data

Install: Paste the contents of STANDALONE-GPT.md into your Custom GPT's Instructions. Recommended Mode: LLM-only (the only available mode). Limitation: No template library, no category loading, no persistence. LLM optimization only. ⚠️ Use STANDALONE-GPT.md, NOT SKILL.md.

---

⭐⭐½ Aider

Feature	Status
Template Engine	⚠️ Can read files but context is tight
LLM Engine	✅
Category Loading	⚠️ Small categories only
State Persistence	❌
Quality Evaluation	✅

Install: Place skill folder in project. Reference with --file. Recommended Mode: "set engine to LLM-only" (saves context space). Limitation: Aider is designed for code editing; general chat feels unnatural. System prompt may be too long with SKILL.md added.

---

📦 Installation Guide

Platform	Files to Use	How to Install
OpenClaw	SKILL.md + all of references/	Place in skills/prompt-optimizer/
Claude Code	SKILL.md + all of references/	Copy to project, reference in CLAUDE.md
Cursor	SKILL.md + all of references/	Copy to project, reference in .cursorrules
Gemini CLI	SKILL.md + small category JSONs	Copy to project, reference in system instructions
Cline	SKILL.md + all of references/	Copy to project, reference in custom instructions
ChatGPT GPT	STANDALONE-GPT.md	Paste into Custom GPT Instructions
Aider	SKILL.md	Copy to project directory

---

🎯 Default Configuration

Setting	Default	Options
Engine	Template-first + LLM fallback	Template-first / LLM-only / Template-only
Library Version	Lite (~3MB, truncated at 800 chars)	Lite / Full (~10MB)
Category	Coding (1,672 prompts, ~1.5MB)	Any combination of 14 categories
Output Format	Plain text	Plain text / Markdown / XML / All
State	Off by default, manually enable	—

---

🚀 Quick Start

Action	Command
Enable	"enable prompt optimizer" / "开启提示词优化"
Enable (Full)	"enable full version" / "开启完整版提示词优化"
Disable	"disable prompt optimizer" / "关闭提示词优化"

Minimal usage:

You: "enable prompt optimizer" (or any language)
You: "optimize this: help me write a login page"
Bot: [outputs optimized prompt]
You: "✅"

---

⚙️ All Configuration Commands

🌐 All commands accept any language. Just express your intent naturally — the AI will understand.

Engine Mode

Command	Behavior
"set engine to template-first" (default)	Search template library first; if no match, auto-fallback to LLM
"set engine to LLM-only"	Skip template library; host AI optimizes directly
"set engine to template-only"	Template library only; tells user if no match found

Template Categories

Command	Behavior
"load only coding templates" (default)	Loads only ~1.5MB
"load coding + data-analysis"	Load multiple categories
"what categories are available"	Show all 14 categories
"load all categories"	Load complete library
"remove art-entertainment"	Remove a loaded category

Category	Prompts	Size
Coding (default)	1,672	~1.5MB
Art & Entertainment	404	~411KB
Other	253	~202KB
Writing	208	~196KB
Education	117	~107KB
Consulting	114	~115KB
Business	108	~98KB
Health & Lifestyle	82	~76KB
Legal & Finance	82	~80KB
Data Analysis	73	~73KB
Creative Generation	73	~66KB
Science & Research	61	~55KB
Translation	60	~57KB
Tech Tools	37	~36KB

Output Format

Command	Effect
"set format to plain text" (default)	Raw prompt text
"set format to Markdown"	Headings, lists, bold formatting
"set format to XML"	<prompt><role>...</role></prompt> structure
"set format to all"	All three formats side by side

---

📋 Post-Optimization Actions

Reply	Action
"✅"	Confirm and use
"❌"	Discard; use original input
Free-form feedback	Fine-tune (e.g., "add responsive design")
"continue optimizing"	Iterate another round
"evaluate"	Show before/after quality scores
"change format"	Switch output format

---

📊 Quality Evaluation

Say "evaluate" (or any equivalent) to trigger a before/after comparison. Scores 1-10 on 5 dimensions:

Dimension	What It Measures
goalClarity	Is the goal clear and specific?
instructionCompleteness	Are instructions complete and unambiguous?
structuralExecutability	Can the structure be executed step by step?
ambiguityControl	Are vague terms eliminated?
robustness	Will it hold up across varied inputs?

Output: Before vs. After scores + total improvement + highlights + remaining gaps.

---

🧠 Four Optimization Modes

Mode	When	What It Does
General (default)	Most scenarios	Structured Role/Profile/Skills/Rules/Workflows output
Analytical	Complex scenarios	8-dimension deep analysis, 5 points per dimension
Iterative	Say "continue optimizing"	Merges new requirements into existing prompt
User Query Refinement	Optimizing user queries	Adds clarity, scope, parameters, output format

---

🔧 Full Workflow

1. Enable → Detect platform capabilities → Load defaults
   ├─ File system available → Template engine ready
   └─ File system unavailable → Auto-switch to LLM-only mode
2. User submits prompt
3. Parse intent (goal, domain, role, constraints, complexity)
4. Engine routing:
   ├─ Template-first: Search → Match → CRAFT output / No match → LLM fallback
   ├─ LLM-only: Meta-prompt generates directly
   └─ Template-only: Search → Match → output / No match → notify user
5. Quality enhancement (precise instructions + constraints + examples)
6. Confirm & deliver
7. Optional: Evaluate / Iterate / Fine-tune

---

🗂️ Template Library Data

Curated from 7 major repositories, 3,344 prompts total:

Source	Description
f/awesome-chatgpt-prompts	CSV, 5MB
awesome-chatgpt-prompts (original)	CSV format
ai-boost/awesome-prompts	228 high-quality detailed prompts
0xeb/TheBigPromptLibrary	115 system prompts
jamesponddotko/llm-prompts	Categorized library
chatgpt-prompts-chinese	Chinese prompts
Prompt Garden	Community contributions

---

📂 File Structure

prompt-optimizer/
├── SKILL.md                          # Core instructions (v4.0, full features)
├── STANDALONE-GPT.md                 # ChatGPT Custom GPT lite version
├── STANDALONE.md                     # Standalone usage guide
├── README.md                         # This file
├── references/
│   ├── prompt_library_full.json      # Full version (~10MB)
│   ├── prompt_library_lite.json      # Lite version (~3MB)
│   └── categories/                   # Split by category
│       ├── index.json
│       ├── coding.json              # Default (coding)
│       └── ... (14 categories)
└── scripts/                          # Build and maintenance scripts

---

🛡️ Features

• Zero API Cost — Host AI optimizes directly, no extra charges
• Auto Platform Detection — Graceful degradation when file system unavailable
• Dual Engine — Template library + LLM meta-prompting
• Category Loading — Load only what you need, save tokens
• Model-Agnostic — Works with all host AI platforms
• Bilingual — Chinese input → Chinese prompt; English → English
• Quality Evaluation — 5-dimension scoring
• Iterative Optimization — Multi-round fine-tuning