Layer 3: Prompts-as-Infrastructure

This directory implements a Block-Based Prompt Architecture designed for Single Source of Truth (SSoT) and scalable maintenance of complex consultant personas.

1. Architecture Overview¶

Instead of monolithic prompt files, the system uses a JIT-composed block architecture. This ensures that shared methodology, technical stacks, and tradecraft are defined once and reused across all consultants.

The SSoT Hub (/blocks)¶

Shared prompt fragments are stored in ai_system_layers/3_prompts/blocks/. These blocks contain authoritative definitions for:

• Common User Stack: The project’s technical environment.

• Core Principles: Behavioral rules and quality standards.

• Consulting Context: Production-ready definitions and industry standards (ISO 29148/SWEBOK).

• Frameworks: Specialized methodologies (e.g., WRC/SVA, Heuer Tradecraft).

• Output Protocols: Standard response structures and formatting.

Consultant Manifests (/consultants)¶

Consultant files are now Manifests—thin skins that describe how to assemble a persona from the hub blocks. A manifest declares its dependencies via an _includes list and provides specific overrides to specialize the persona.

JIT Compilation¶

The prepare_prompt.py script acts as the compiler. At runtime, it:

1. Recursively resolves all _includes.

2. Merges the blocks into a single structure.

3. Applies overrides from the manifest (Manifest values > Block values).

4. Converts the resulting JSON to a clean, YAML-like format for LLM delivery.

2. Directory Structure¶

3. Creating a New Consultant¶

To create a new consultant, define a JSON manifest in /consultants following this pattern:

{
  "metadata": {
    "title": "Specialized Consultant",
    "version": "1.0.0"
  },
  "_includes": [
    "blocks/common_principles.json",
    "blocks/common_user_stack.json",
    "blocks/standard_output_protocol.json"
  ],
  "consultant_persona": {
    "role": "Expert in X",
    "tone": "Analytical"
  },
  "USER_INPUT": "..."
}