AI Prompt Optimizer

The CO-STAR Framework

Created for business and complex writing tasks. CO-STAR ensures the LLM understands all operational parameters before writing.

• C - Context: Background info & scenario.
• O - Objective: The precise end deliverable.
• S - Style: Writing style (e.g. Steve Jobs, Technical).
• T - Tone: Emotional resonance & formality level.
• A - Audience: Target reader or consumer.
• R - Response Format: Markdown, JSON, table.

Structural XML Tagging

Recommended officially by Anthropic for Claude and OpenAI for GPT-4o. XML tags like <context> and <instructions> isolate inputs from instructions.

This prevents prompt injection, ensures precise parsing of datasets, and enforces strict boundary adherence across long context windows.

Role & Persona Priming

Assigning a specific domain persona (e.g. "Act as a Principal Staff Engineer with 15+ years of distributed systems experience") shifts the model's latent weights toward specialized domain terminology and industry standard patterns.

Negative Constraints

Telling an AI what NOT to do is just as critical as telling it what to do. Explicitly prohibiting buzzwords (e.g. "delve", "testament", "tapestry"), speculation, or excessive preamble prevents generic AI tropes.