Prompt Architecture

# Prompt Architecture

Part of [Agent Skills™](https://github.com/itallstartedwithaidea/agent-skills) by [googleadsagent.ai™](https://googleadsagent.ai)

## Description

Prompt Architecture is the structural engineering of agent instructions. Where casual prompt writing produces fragile, inconsistent results, architectural prompt design creates deterministic, high-performance agent behaviors that hold up under adversarial conditions and scale across thousands of invocations. This skill distills the prompt engineering methodology developed within the [googleadsagent.ai™](https://googleadsagent.ai) platform, where Buddy™ handles complex Google Ads analysis through meticulously layered prompt structures.

The fundamental principle is that prompts are not strings — they are programs. A well-architected prompt has a clear execution model: system-level invariants establish the agent's identity and constraints, user-level instructions define the current task, and assistant-level priming shapes the output format and reasoning trajectory. Each layer serves a distinct purpose and must be engineered independently before composition.

Advanced prompt architecture incorporates constraint propagation, output schema enforcement, chain-of-thought scaffolding, and dynamic few-shot example selection. These techniques eliminate the "prompt lottery" problem where identical inputs produce wildly varying output quality across runs.

## Use When

- Agent outputs are inconsistent across invocations with the same input
- You need deterministic formatting (JSON, structured reports, specific schemas)
- Complex multi-step reasoning requires explicit chain-of-thought scaffolding
- The agent must adhere to strict behavioral constraints (safety, tone, scope)
- Few-shot examples are needed to establish domain-specific patterns
- You are designing system prompts for production deployment at scale

## How It Works

```mermaid
graph TD
    A[System Layer] --> B[Identity & Constraints]
    A --> C[Output Schema Definition]
    A --> D[Tool Definitions]
    B --> E[Prompt Assembly]
    C --> E
    D --> E
    F[User Layer] --> G[Task Specification]
    F --> H[Dynamic Few-Shot Examples]
    G --> E
    H --> E
    I[Assistant Layer] --> J[Reasoning Primer]
    I --> K[Format Enforcement]
    J --> E
    K --> E
    E --> L[Validation Gate]
    L -->|Pass| M[Agent Execution]
    L -->|Fail| N[Prompt Revision]
    N --> E
```

The three-layer architecture ensures separation of concerns. The system layer defines who the agent is and what it can do — this layer rarely changes across invocations. The user layer carries the task-specific payload and any dynamically selected examples. The assistant layer provides a "running start" that primes the model's generation trajectory. The validation gate checks assembled prompts against structural rules before execution, catching malformed or conflicting instructions.

## Implementation

**Three-Layer Prompt Builder:**

```typescript
interface PromptLayer {
  role: "system" | "user" | "assistant";
  sections: PromptSection[];
}

interface PromptSection {
  name: string;
  content: string;
  priority: number;
  tokenBudget: number;
}

function assemblePrompt(layers: PromptLayer[], maxTokens: number): Message[] {
  const messages: Message[] = [];
  for (const layer of layers) {
    const sections = layer.sections
      .sort((a, b) => b.priority - a.priority)
      .reduce((acc, section) => {
        const currentTokens = countTokens(acc.map(s => s.content).join("\n"));
        if (currentTokens + section.tokenBudget <= maxTokens * 0.4) {
          acc.push(section);
        }
        return acc;
      }, [] as PromptSection[]);

    messages.push({
      role: layer.role,
      content: sections.map(s => s.content).join("\n\n"),
    });
  }
  return messages;
}
```

**Constrained Output Enforcement:**

``