Side questions — /btw

Ask about a flow without interrupting it. /btw opens a side conversation beside the run, with an agent that can read the run but not change it. The flow's own agents never see what you ask there.

At a glance

• You will ask how a run is going, what an agent is doing, or why it did something, and get an answer in words.
• Use it when the transcript is too long to read, or you want an explanation rather than a change of course.
• You need a run in hmz, running or ended. Your first run starts one.

Try it

While a flow is running, or after it has ended, type:

❯ /btw what is the reviewer waiting for?

Press esc when you are done.

How it works

A line you type into a running flow steers it: the agent hears it and acts on it. A side question must never do that, so /btw sends it somewhere else: to a side conversation that runs read-only, next to the run, and is closed when you leave.

/btw puts you in btw mode. From then on every line you type is one more question in the same side conversation, so a follow-up can say "and then?". The status line starts with btw · <who> for as long as you are in it.

Who answers depends on what is on the screen when you type /btw:

• An agent's or one conversation's transcript (btw · NeiKos496/2): a side copy of that conversation, the agent's newest. Where its CLI can fork a conversation (Claude Code, Codex, Grok Build, Kimi Code, MiMo Code, Oh My Pi, opencode, pi, Qwen Code, and a CLI you added that speaks ACP), the copy is a fork and knows everything the conversation knew. Otherwise, or if the fork fails, it is a fresh session of the same agent, given a snapshot of the run and what that role has done lately. This works for a conversation that has ended too, with no flow running.

• The monitor, or the screen before a run opens a conversation (btw · The Epic): The Epic, the agent chosen as /btw The Epic on the General page of /settings -- the same one for every epic. With none chosen there, nobody answers The Epic: no agent of the run stands in. It is given a snapshot of the run, and with every question the list of its sessions, and it may put a question of its own to any session's side copy, up to 4 per question you ask. Each one shows as a dim btw · asking <session>: … line.

  The Epic carries an ask-conversation skill that tells it to ask a session rather than guess, by running hmz-ask ask --to <session> --question "…" (hmz-ask list lists the sessions). hmz catches that command where The Epic runs and answers it mid-turn: Claude Code and Qwen Code hear the answer through a hook, a CLI that ran the command reads it from its output, and any other CLI is told it when its turn would end. A CLI that cannot run a command at the read-only rung asks by replying with @ask <session>: <question> lines instead. MiniMax Code, DeepSeek Harness and ACP agents have no read-only rung, so they cannot answer The Epic; a side question to one fails, saying so.

The snapshot holds:

• the flow, its task, and how long it has been going;
• each agent's model, how many turns it has taken, and whether it is working;
• the handovers between agents, and what the run has spent;
• the latest things the agents said and did, and what the flow printed.

Example: ask what the reviewer is waiting for

An rlar run has been going a few minutes. The actor is working and the reviewer is idle, and you want to know why. You are reading the actor, where the run landed, so a side copy of its conversation answers. Step through it:

hmz · rlardrawn, not recorded

0000000000

─ reading actor ─

● actor is working

❯ /btw what is the reviewer waiting for?

btw · actor/11 each line is a question; /btw or esc to exit2

actor · claude/claude-opus-5-5:high · ● 1

reviewer · codex/gpt-5.6-sol:high · ○ 1

❯

btw · actor/13 · ·|· actor… (51s · ctrl+c twice to stop)shift+tab switch view · / commands · shift+enter newline · ← monitor · ctrl+c stop

The run carries on. The actor never sees the question, and the status line says you are in btw mode.

What to look at, by number:

1. btw · actor/1. Who is answering: a side copy of the conversation you are reading. On the monitor this is btw · The Epic instead.
2. each line is a question; /btw or esc to exit. You are in btw mode now. A plain line typed from here is a question, not a steer.
3. The status line starts with btw · actor/1 until you leave, so you cannot forget you are in it.
4. The answer, in cyan, behind your question in dim. It lands in the transcript you are reading, among the run's own lines, and nothing about the run changed.
5. btw: exited. esc, or /btw on its own, leaves btw mode and closes the side conversation.

Check it worked

• The actor's ● 1 line is unchanged: it never saw the question.
• The answer is cyan and starts with btw ·, which no line of the run's own does.
• Once you leave, the status line no longer starts with btw.

Variations

• Ask about one conversation. Press ← for the monitor, walk to a session and press enter to read it, then type /btw. A fork of that conversation answers, so it remembers its own reasoning:

  ❯ /btw what did you say, in three words?
  btw · assistant/1 each line is a question; /btw or esc to exit
  ● btw · what did you say, in three words? Build something great.

• Enter first, ask after. /btw on its own enters btw mode with no question yet.

• Choose who answers The Epic. Set /btw The Epic on the General page of /settings (on the web page, The Epic under Settings › General): none, the one chosen, or another…, which asks for its CLI, account, model and effort. It applies to every epic, from the next side conversation of The Epic on. A conversation's side copy is always that conversation's own agent: there is nothing to pick for it.

• On the web page, the Aside tab keeps several side conversations at once, one tab each, and New aside opens another. A new one's Ask chip, above its box, picks whom it asks: The Epic, then each agent with the role it fills, opening onto its conversations, newest first. Type in the box at its top to narrow it to the agents, roles and conversations that match. Its first question fixes whom it asks. While it answers it says so, with each conversation The Epic asks on the way.

What a side conversation may do

Every side conversation runs read-only, with no skills, no goals and none of the flow's tools. It can read the workspace, but it cannot change anything and it cannot steer the flow. Its turns are real turns, though: they count against the run's budget. They do not appear on the monitor or in the run's trace.

Starting a new flow, or closing hmz, also leaves btw mode.

Troubleshooting

hmz: /btw: Nobody answers The Epic yet: set the agent that does in the settings, under General.

No agent is chosen to answer The Epic. Choose one as /btw The Epic in /settings (or The Epic under Settings › General on the web page), or read a conversation and ask that instead.

hmz: /btw: no conversation found for <session>

The conversation on the screen is gone, so there is nothing to copy. Read another one, or go up to the monitor and ask The Epic.

hmz: btw is still answering the last question

Questions go one at a time. Wait for the answer, then ask again.

hmz: /btw: …

The side question failed, for the reason given. The flow is not affected.

The agent did what I asked in btw mode

It cannot: nothing in btw mode reaches the flow. If the agent changed course, the line was typed outside btw mode. Check the status line starts with btw before you type.

Next steps

• Talking to a running turn: tell the agent something rather than ask.
• Watching a run: see the run's shape without asking.
• Many conversations at once: choose whose side copy answers.
• TUI reference: every command and key.