# Start here

**If you are not a developer.** UCC is the setup ZT uses to run agents. An agent is a program on your computer that reads and writes files, keeps notes in a folder that is backed up, builds web pages you can share, and reaches you on Telegram when it needs an answer. Step 0 checks you have ZT's message; steps 1 to 5 set everything up. Each step has one short paragraph, a text to paste to your agent, and a line saying how to check it worked. If a step does not pass, send ZT the step number and what your agent printed.

**If you are technical.** Five steps on one machine. At the end you have one Claude Code agent the daemon knows by name, a sessions root where everything it writes survives, a bot that reaches you on Telegram, and a wiki your agents grow — with nothing to babysit.

## What the steps produce

| Step | You end up with | Proof |
|---|---|---|
| 0 Before you start | The message from ZT, in hand: this handbook's link, the install line with its three values, the invitation to the private repositories on git.0xdao.app | the message itself |
| 1 Install | UCC and the `ccc-statusd` daemon on your machine, already running | `ccc-statusd status` |
| 2 First session | One Claude Code agent with an eight-character name and a card of its own | the `joined … as agent …` banner |
| 3 Sessions root | The agent's records pushed to your own private GitHub repository every ten minutes | `bin/check` prints `check: ok` |
| 4 Telegram | Your own bot in a group of your own, so an agent can ask you a question from anywhere | a test question reaches your group with buttons |
| 5 Wiki | A wiki that grows from what your agents learn | `mrd run LLM_WIKI.md check` passes |

## Who "your agent" is in steps 0, 1 and 2

Your UCC agent does not exist until step 2. Until then, "your agent" is whatever AI assistant you already use. Step 0 needs only a chat. Steps 1 and 2 need an assistant that can run commands in a terminal on this computer, such as Claude Code or Codex; if you have none, do those two steps yourself — each is one line typed into the Terminal app, and the step says which. From step 3 on, "your agent" is the UCC agent step 2 started, in the window it opened.

## The order

```mermaid
flowchart LR
  s1[Install] --> s2[First session] --> s3[Sessions root]
  s3 --> s4[Telegram]
  s3 --> s5[Wiki]
```

Do the steps in number order. When step 5 passes, the closing card at the end of this page says where to go next.

## Dive in: why this order

Install first: everything after it is a file the installer put on your machine. The first session second: it proves the daemon and its hooks fire before anything depends on them. The sessions root third: from the first session on, every card, memo and result an agent writes lands in that folder. The installer creates it, but it is not durable until step 3 names your repository, and the daemon's sweep reports red every cycle until then. Telegram and the wiki each need only the root; do Telegram first — an agent that cannot reach you stalls on its first question — then the wiki, which graduates from the root.

## Dive in: what comes from ZT, and only from ZT

One direct message from ZT carries everything this page cannot: the link to this handbook; the install line, already assembled from your three install values — the username, the token and the `ENCRYPTION_PASSWORD`; and the invitation to ZT's private code repositories on git.0xdao.app. None of the five steps needs the repositories; the [Repos](#repos) section lists them. Nothing on this page carries a secret, and nothing on it asks you to paste one anywhere except your own terminal, once, in step 1.

## Dive in: how a step reads

Every step reads the same way: its title, one short paragraph saying what the step does, then a **Tell your agent** panel. Its **Copy the prompt** button copies the whole prompt your agent needs; the last copied line names the web address the prompt lives at, so an agent that can read web addresses re-reads the current text. Under the panel, **How to check it worked** is what you or your agent run and what a pass prints. The full prompt is folded under **The full prompt your agent reads**, and the **Skill** line under it names what the agent loads for the step.