Prompt Optimizer Codex Skill

Prompt Optimizer Codex is a reusable skill package that transforms vague requests into execution-ready prompts and can optionally execute them end-to-end.

Core methods:

• Lightweight grounding research
• Targeted clarification questions
• EARS (Easy Approach to Requirements Syntax) requirement rewriting
• DevBooks-style phased routing and acceptance gates
• Autonomous execution loop with bounded retries

Repository Structure

• SKILL.md: Main skill contract and workflow
• references/ears-cheatsheet.md: EARS patterns and rewrite checklist
• references/clarification-playbook.md: Clarification-question strategy
• references/devbooks-integration-patterns.md: Routing and gate patterns borrowed from spec-coding workflows
• references/high-quality-prompt-patterns.md: Extracted benchmark patterns from high-performing engineering prompts
• agents/openai.yaml: Agent metadata

Installation

1. Clone this repository.
2. Copy the folder into your Codex skills directory.
3. Trigger it in chat with:

$prompt-optimizer-codex 优化下这个 prompt

Modes

1. optimize-only

• Output optimized prompt package only.

2. optimize-and-run

• Output optimized prompt package and execute it with an evidence-based report.

Usage

The skill outputs, in order:

1. Original Request Summary
2. Gaps Detected
3. Clarification Questions (only when unresolved decisions remain)
4. EARS Requirements
5. Optimized Prompt
6. Assumptions and Risks
7. Execution Report (only in run mode)

Quality Guarantees

• Requirements are testable and atomic.
• Conditions and thresholds are explicit.
• Ambiguous language is converted into measurable constraints.
• In run mode, each step includes acceptance evidence.

Development

Run local validation:

pwsh ./scripts/validate.ps1

Roadmap

• Add domain-specific prompt profiles
• Add bilingual output presets
• Add automated requirement-quality checks
• Add reusable execution report templates for common workflows

License

MIT