Kitchen Metaphor

PREQSTATION is named after a kitchen prep station — a place where a chef directs work and line cooks prepare ingredients. This metaphor guides how the system works.

The Kitchen Metaphor

Kitchen Concept	PREQSTATION Equivalent	Meaning
Chef	Developer (you)	Directs tasks, makes final decisions
Line Cook	AI Agent	Executes tasks (Claude Code, Codex, Gemini)
Ticket	Task	A unit of work with acceptance criteria
Queue	Task Queue	Waiting tasks in status order
Station	Agent Instance	An AI working on a specific task
Prep List	Acceptance Criteria	Instructions for completing the task
Mise en Place	Setup	Configuring API tokens and environment
Fire	Execute	Launch a task for an agent to work on
Pass	Ready	Work is complete and waiting for verification
Order Up	Done	Task verified and finished
86’d	Hold	Task cannot proceed until blocker is resolved

The Story

In a kitchen, the head chef doesn’t cut onions or cook steaks. Instead, the chef:

1. Receives orders — understands what the customer wants
2. Writes tickets — creates clear instructions with acceptance criteria
3. Directs line cooks — tells them which station to work on
4. Inspects results — checks quality before service
5. Serves the customer — ensures satisfaction

PREQSTATION works the same way:

1. You write tasks — describe what you want the AI to implement
2. Set acceptance criteria — list what “done” means
3. AI agents execute — they plan, implement code, run tests, and move work to ready
4. You review — verify the result, then approve to done or return it to hold
5. Deploy — merge and ship

Note

You are the chef. The AI agent is the line cook. Your job is to direct and inspect, not to do all the prep work yourself.

Key Principles

1. Clear Tickets

A good task is like a good kitchen ticket:

• Specific objective
• Clear acceptance criteria (checklist)
• Priority level
• Any constraints or context

2. Isolated Worktrees

Just like each station in a kitchen is separate, each agent runs in an isolated git worktree. No interference, no pollution of your main branch.

3. Status Visibility

The Kanban board shows all tasks in their current state:

• Inbox — New tasks waiting for planning
• Todo — Planned and ready for agents
• Hold — Blocked or paused; needs more context or retry
• Ready — Implementation complete, waiting for verification
• Done — Task verified and complete

Execution overlays are separate from workflow columns:

• Queued — Requested or dispatched, waiting for pickup
• Working — Claimed by an agent right now

4. Asynchronous Execution

Agents run independently. You don’t watch them work. You check results periodically.

Terminology

Term	Meaning
Task	A unit of work with acceptance criteria and status
Engine	Which AI agent will execute: claude-code, codex, or gemini-cli
Status	Workflow state: inbox, todo, hold, ready, done, archived
Run State	Live execution overlay: queued, working, or null
Acceptance Criteria	Checklist of what must be true for task completion
Work Log	Result of agent execution: summary, tests, PR URL, timestamp
API Token	Bearer token for agents to authenticate to REST API (prefix: preq_)

When to Use PREQSTATION

Good use cases:

• Refactoring existing code
• Adding features with clear specs
• Writing tests for untested code
• Implementing API endpoints
• Fixing bugs with reproduction steps
• Deploying to production
• Updating dependencies

Not ideal:

• Exploratory work (needs human judgment)
• Architectural decisions (needs discussion first)
• Creative writing or design
• Work requiring deep context about your business logic