Skip to content

Finish your first task in the terminal ​

In ten minutes, you install hmz, chat with a coding agent, install a flow, and run it on a task in your project. Everything happens in one terminal.

The terminal interface as it opens: a banner, the agent it will talk to, and a promptThe terminal interface as it opens: a banner, the agent it will talk to, and a prompt

What you need ​

  • A terminal on macOS or Linux, or a WSL terminal on Windows.
  • uv. Install it with curl -LsSf https://astral.sh/uv/install.sh | sh.
  • A project directory. This page uses ~/shop.
  • An account with one coding agent, such as Claude Code, and npm to install its CLI. Humanize runs that CLI under your login.

1. Install the CLI ​

Install hmz:

sh
uv tool install hmz

Verify the installation:

sh
hmz --version

hmz --version prints the versionhmz --version prints the version

2. Sign in to a coding agent ​

Humanize drives a coding agent CLI, called a harness. You need one. To set up Claude Code, install it and sign in:

sh
npm i -g @anthropic-ai/claude-code
claude auth login

Skip this step if you already have a coding agent CLI on this machine. See Coding agents for other harnesses.

3. Open hmz in your project ​

Open your project directory and start hmz:

sh
cd ~/shop
hmz

The terminal interface opens on a new chat. The line above the prompt shows the active agent: the harness, model, and effort. On first run, a box asks Report errors to Humanize?. Answer yes or no.

If the line says no agent installed, return to step 2. To switch agents, type /settings workspace and open Chat agent.

4. Chat with the agent ​

Type a question and press enter:

text
Where is checkout handled?

The agent answers. A chat needs no flow or budget.

A question is typed, the agent works, and its answer appearsA question is typed, the agent works, and its answer appears

5. Install a flow ​

A flow is a loop that runs agents on a task. Humanize starts with no flows installed. You install them from a flowverse, and the official flowverse is already listed.

  1. Type /flow and press enter. The Installed page is empty and Install more… is highlighted.
  2. Press enter. The Flowverses page lists official.
  3. Press enter to view its flows.
  4. Select ralph_loop and press enter to see its releases.
  5. Press enter on the newest release. It shows ✔ installed.
  6. Press esc until the menu closes.

The flow menu: Install more, the official flowverse, ralph_loop, its release, installedThe flow menu: Install more, the official flowverse, ralph_loop, its release, installed

ralph_loop gives one agent the same task in a fresh session every round until its budget runs out or you stop it. The catalogue describes the other flows.

TIP

If official says not fetched, press tab, select Fetch, and press enter. hmz fetches this index in the background each time it opens, so this only happens when that fetch has not finished or failed.

6. Give the flow a budget ​

Every flow run requires a budget to know when to stop.

  1. Type /flow ralph_loop and press enter. The page lists the flow's settings. The agent row already names an agent; press enter on it to choose another.
  2. Select budget and press enter.
  3. On duration, press enter, type 10m, and press enter again.
  4. Press tab to reach Set, then press enter. The budget row now reads stops at 10m.
  5. Press esc. A box asks Save?. Press enter on Save.

The budget is opened, set to ten minutes, and savedThe budget is opened, set to ten minutes, and saved

The transcript shows set up: /flow ralph_loop <task> starts it. A budget can also limit cost in US dollars and output_tokens. The run stops at the first limit it reaches.

7. Start it on a task ​

Run the flow on a task:

text
/flow ralph_loop Add a discount code to checkout

The screen switches from the chat to the flow's agent. Each round appears as the agent finishes it. This recorded run of a flow is called an epic.

The flow's agent at work on the task, round by roundThe flow's agent at work on the task, round by round

8. Watch the monitor ​

With the prompt empty, press ←. The monitor displays the epic: its state, turns taken, budget time used, a box per agent, and tokens spent.

The monitor: the epic's state and budget, the agent's box, and the tokens it has usedThe monitor: the epic's state and budget, the agent's box, and the tokens it has used

Press → to return to the agent's transcript.

9. Stop the run ​

Left alone, the run stops after ten minutes once the current turn finishes. To stop it now, press ctrl+c. The transcript says — press ctrl+c again to stop the flow —. Press ctrl+c again within three seconds. Typing /stop stops the run immediately without confirmation.

The transcript after ctrl+c twice: the flow is stopping and the open turn is cut offThe transcript after ctrl+c twice: the flow is stopping and the open turn is cut off

The agent's changes stay in your project.

10. Read the result ​

  1. Type /epics and press enter. Every epic for this directory appears, newest first, with its flow, task, duration, and status.
  2. Press enter on the epic, then enter on graph.
  3. Press enter on a session to read the task, each tool call and output, and the agent's response.

The list of epics, with the stopped ralph_loop runThe list of epics, with the stopped ralph_loop run

One session read back: the task, a tool call and its output, and the agent's answerOne session read back: the task, a tool call and its output, and the agent's answer

Press esc to back out of each page. Type /exit to leave hmz.

What you did ​

  • Installed hmz and chatted with a coding agent.
  • Installed ralph_loop from the official flowverse.
  • Set a budget and started an epic on a task.
  • Watched the epic in the monitor and stopped it.
  • Read the session history in /epics.

Next ​

Open source on github.com/humanfia · RSS: blog · news