Skip to content

Agent Fundamentals

How a terminal agent, a kanban board, and a handful of plain files run a real software project — skills, tools, tool specs, MCPs, Speccify, and iKanban AI.

A software project keeps its own operating manual inside the repository: a .agent/ folder with a plan, a kanban board, skills, and tools — all plain markdown and JSON. A terminal agent (Claude Code or any other) reads that folder and works the board; the owner watches and steers through the iKanban AI app, which renders the very same files. No hidden database, no proprietary format: everything the agent knows, you can open in a text editor.

The tutorial

4Notice, zero to App Store: one real app built with plans, tickets, atomic commits, skills, fastlane, and App Store Connect automation — the site’s main act.

Speccify

Expand, execute, evaluate: turning a source of skills and tools into checked, working implementations.

iKanban AI

Working with the file-based kanban board: plans, tickets, history, and the question protocol between owner and agent.

Practiced, not theorized

Every example on this site is lifted from a real project that runs this workflow every day — including this site itself.