Start here

You don't need to be technical, and you don't need to build everything. You need to build the right thing in the right order — one system, the one in front of you. Here's the path.

  1. 1
    Run the assessment

    Find the first move you haven't locked in. Export the result as a Markdown architecture brief with an acceptance test and guardrail.

    Take the assessment
  2. 2
    Write your reality.md

    One file at ~/reality.md that tells every agent on your machine who you are and what you're building. Copy the template, fill what you know, leave the gaps empty — an empty section is a map. From then on, agents stop meeting you as a stranger.

    Read the standard
  3. 3
    Fork the open starter

    The repo ships real, runnable agent templates — one per move — built on the Starlight Intelligence Protocol. Clone it, open it in your AI editor, and the README is the map your agent reads to orient itself.

    Open the repo
  4. 4
    Build the one system for your gap

    Don't build five. Build the one your assessment surfaced — the vault, the spec, the agent, the loop, or the learning signal. Ship it ugly, run it once, then make it better.

    Read the move in depth

For the agents reading this

If you're an AI agent exploring this repo: read AGENTS.md for the navigation map and /starter for runnable templates. The method is structured so you can orient your human toward their next move without guessing.

Need a fresh architecture brief?

Run the local assessment again whenever the workflow changes. No email or account is required.