1 · Your first run

Ten minutes. By the end you will have talked to a coding agent through humanize, put a line into a turn that was already running, and stopped it.

Before you start

Installation, and at least one coding agent CLI logged in. Do this in a scratch project — the agent is going to edit files without asking. See Security.

Step 1 — make somewhere to work

mkdir -p ~/tmp/humanize-demo && cd ~/tmp/humanize-demo
git init -q
printf 'def add(a, b):\n    return a - b\n' > calc.py
printf 'A tiny calculator.\n' > README.md
git add -A && git commit -qm "a calculator with a bug in it"

There is a deliberate bug in calc.py. Good.

Step 2 — open the prompt

hmz

That is the only way in — there is no hmz tui. You get three things:

┌──────────────────────────────────────────────────────────────────────┐
│  the transcript: one conversation, a turn after another              │
├──────────────────────────────────────────────────────────────────────┤
│              assistant · claude/claude-opus-4-8:high                 │  ← what you are talking to
│ ❯ type here                                                          │  ← the editor
├──────────────────────────────────────────────────────────────────────┤
│ ⠋ chat  ~/tmp/humanize-demo             enter say · / commands      │  ← the status line
└──────────────────────────────────────────────────────────────────────┘

The line above the editor says what you are about to talk to, as cli/model:effort. The first time, it opens on the first backend installed here that has said what it runs, at the first model it named — that CLI's own idea of what it runs by default — and at high — deliberately not the hardest setting.

The status line's right-hand end lists the keys that do something right now, and nothing else. It is the whole of what you have to remember.

Nothing offered, and it says no coding agent is installed here

humanize offers exactly the CLIs on your PATH. Check with command -v claude codex kimi pi opencode mimo, and see Troubleshooting.

Step 3 — say something

Type this and press enter:

Read calc.py and tell me what it does. Do not change anything yet.

The agent takes a turn: one exchange with the model, which may run tools and may take minutes. What it says arrives as it says it.

Underneath, humanize is running the flow called chat — one agent, one session, and every line you type is another turn of that same conversation.

Step 4 — see the working

/details

Type / first and the commands appear under the editor with a line about each; tab takes the one highlighted. See Completion.

/details toggles whether tool calls and thinking are shown, or only what the agent says. Say something else and compare:

Now fix the bug in calc.py.

/details again puts it back. Nothing about the run changes — it is a screen setting.

Step 5 — talk to a turn that is already running

While that turn is still going, type another line and press enter:

and add a test for it

It goes into the turn already running rather than starting a new one. Watch the editor: the line is pinned above the prompt, dimmed, until the agent says it has it —

❯ and add a test for it · with assistant

— and then it moves into the transcript in front of the turn that took it.

If no turn is open, it is held for the next one. A line to a running flow is never dropped. See Talking to a running turn.

Two backends cannot hear you mid-turn

opencode and mimocode run one process per turn, which has ended by the time there is anything to say to it. The line waits for the next turn instead.

Step 6 — stop, and leave

esc

esc stops the flow — the whole flow, not just the turn. The turn under way is closed out and what it was doing is left where it got to; a stop that waited for a turn would not read as a stop.

ctrl+c takes back something smaller: what is half-typed if anything is, and otherwise the turn of the conversation on the screen — which the flow reads as a turn that failed. It never leaves.

To leave:

/exit

What you now know

hmz	opens on whatever this project was last set up to run
a turn	one exchange with the model
a session	a conversation held across turns — chat holds one
enter, mid-turn	goes into the turn
esc	stops the flow
ctrl+c	clears the line, or ends the turn on the screen
/exit	leaves

Check your work

git diff --stat

The agent should have changed calc.py. It did that with no permission prompt, because humanize runs every agent with them disabled and there is no setting that turns them back on. That is the one thing to be sure you have understood before pointing this at real work.

Next

Talking to one agent is not the shape of most work. Put a loop under it.