docmd is architected as an “AI-First” documentation engine. By adhering to these structural best practices, you ensure that LLMs (such as ChatGPT, Claude, and GitHub Copilot) can parse your project’s logic and architecture with surgical precision.

1. Enable the LLM Plugin

The baseline for AI optimization is the native llms plugin. It generates structured context files specifically designed for model ingestion.

// docmd.config.js
export default {
  plugins: {
    llms: { 
      fullContext: true // Generates the comprehensive llms-full.txt
    }
  }
}

2. Semantic Heading Integrity

AI models utilize H-tags to build a hierarchical map of internal technical relationships.

• Logical Descent: Never skip heading levels (always go H1 → H2 → H3).
• Technical Density: Use descriptive headings. Instead of “Auth,” use “Implementing OAuth2 Password Grants.”
• The H1 Singular: Ensure your Markdown frontmatter title is descriptive; docmd utilizes this as the primary semantic entry point.

3. Lexical Code Metadata

Always explicitly specify the language identifier for fenced code blocks. This allows the AI’s internal tokenizer to apply the correct grammar rules during context retrieval.

```typescript
// Optimized entry point
const docmd = new Engine();
```

4. Leveraging the Context Pipeline

The llms-full.txt file is a high-fidelity, concatenated stream of your entire static site.

• Prompt Engineering: Direct your AI: “Use the semantic structure in /llms.txt and the comprehensive technical content in /llms-full.txt to analyze this codebase.”
• Context Control: Use llms: false in specific page frontmatter to exclude sensitive or internal-only documentation from the public AI context stream.

5. High-Fidelity Alt-Text

While vision-capable models (Multimodal LLMs) are advancing, descriptive text remains the most reliable signal for reasoning engines. Comprehensive alt text for diagrams and screenshots ensures that the agent understands the visual logic even during text-only processing phases.