Skip to main content

The Agent System

specs.md provides four specialized agents that guide you through the entire development lifecycle. Each agent has a focused responsibility and set of commands.

Agent Architecture

Agent Responsibilities

Master Agent

Orchestrates the overall flow, routes requests, maintains project awareness

Inception Agent

Captures intents, elaborates requirements, decomposes into units

Construction Agent

Executes bolts, builds code, runs tests through validated stages

Operations Agent

Deploys, verifies, and monitors systems in production

Invoking Agents

These are NOT command-line commands. These are prompts you type in your AI coding tool’s chat interface (Claude Code, Cursor, Copilot, etc.).

How to Use

  1. Open your AI coding tool (Claude Code, Cursor, GitHub Copilot)
  2. Start a chat/conversation
  3. Type the agent prompt to activate it
Type in Claude Code’s chat:

Agent Session Lifecycle

Each agent session follows a pattern:
1

Context Loading

Agent reads relevant artifacts from Memory Bank
2

Command Execution

Agent performs the requested command
3

Human Validation

Agent pauses for approval at checkpoints
4

Artifact Storage

Agent writes results back to Memory Bank
Each agent invocation starts fresh. Agents read context from the Memory Bank at startup. Ensure artifacts are saved after each step.

Command Reference

Master Agent

Inception Agent

Construction Agent

Operations Agent

Routing Between Agents

The Master Agent can route you to the appropriate agent:

Best Practices

When unsure where to begin, start with the Master Agent. It will guide you to the right place.
Finish and save artifacts before switching agents. Context is loaded from Memory Bank.
Inception → Construction → Operations. Don’t skip phases.
Each agent has specific commands. Use them rather than free-form requests.