How-to · human-in-the-loop

Approve agent actions from your phone

Q: Can I approve a coding agent's action from my phone?

Yes. Instead of a terminal y/n prompt, your agent pushes a Meltbox brief; you get a push on your phone and pick an option from the reader. The selection flows straight back to the agent as its next step, so a decision made on your phone continues the run.

The pattern that burns everyone: you ask Claude Code to do something, it stops at a y/n prompt in the terminal, and now the whole run is blocked on you being at that desk. Here's how to make the approval land on your phone — and count.

Create a free workspace See the live demo

Short answer. Instead of a terminal y/n, have your agent push a Meltbox brief — a page with real choices. It arrives as a push on your phone; you pick an option from the reader and the selection flows back as the agent's next step. A decision made on your phone continues the run.

The problem

A terminal prompt is a leash

Agents that act too fast get you burned, so you keep a human gate: the agent asks before it does the risky thing. But the standard gate is a y/n in the terminal — which means the agent is idle until you're back at the keyboard. The common workarounds don't really fix it:

A Slack ping tells you something's waiting, but it has no buttons the agent reads — you still SSH in or wait for your desk to answer.
SSH from your phone works if you enjoy a tmux pane on a 6-inch screen. Most people don't.
Auto-approve everything removes the gate — which is the thing that burned you.

"Claude Code already asks me — isn't that the gate?"

It is a gate — and that's exactly where the pain moved. Once agents do most of the typing, the bottleneck isn't writing code, it's the stream of Continue? [y/N] prompts you approve without really reading. A terminal that can only ask yes or no, right here, right now trains you to rubber-stamp. The fix isn't a better prompt — it's a real decision surface (options, context, a note) you can answer deliberately from wherever you are.

The fix

Make the approval a brief

A brief is a self-contained page with clickable choices and a note box. The agent pushes it and keeps a monitor running; you get a push, open it on your phone, and answer. Your selection becomes an event the agent reads as its next instruction — so the phone is where the decision happens, not just where you're alerted.

the agent asks — as a real choice, not a y/n

mb brief push ./approve.html \
  --title "Run the migration on prod?" --kind options
mb brief watch br_8f3k --since 0 --interval 3
# choices in approve.html, e.g.:
#   [Run it now]  [Dry-run first]  [Hold — I'll look tonight]

you decide — from the lock screen

[push] Your agent needs a decision — "Run the migration on prod?"
# tap → the brief opens → pick "Dry-run first" → the agent picks it up and proceeds.

On mobile you can open the brief, click a choice (recorded instantly), leave a note (n), or archive to signal you've reviewed and decided. The reader is keyboard- and touch-fast, and works on Mac and iPhone with push even when the app is closed. If the decision needs more than a tap, the note box carries free-form direction back to the agent.

Prefer to decide in chat?

If you connect Meltbox to ChatGPT or Claude over its MCP server, the brief renders as a card in the conversation — and just saying "go with the dry-run" records the same selection. Same loop, wherever you are.

Side by side

Ways to approve an agent action remotely

	Terminal y/n	Slack ping	Meltbox brief
Reaches your phone	No	Yes	Yes
You can answer on your phone	No	No	Yes
More than yes/no (options + note)	No	No	Yes
Answer feeds the agent	at the desk	No	Yes
Keeps a real gate	Yes	sort of	Yes

Meltbox flow verified against Authoring briefs, Quickstart, and MCP server, July 2026.

FAQ

Common questions

Can I approve a coding agent's action from my phone?

Yes. Instead of a terminal y/n, your agent pushes a Meltbox brief; you get a push and pick an option from the reader. The selection flows straight back to the agent as its next step, so a decision made on your phone continues the run.

Is a Slack notification enough?

A Slack ping tells you something is waiting but has no buttons the agent reads — you still SSH in or wait for your desk. A brief carries the actual choices and the reply loop, so the phone is where you decide, not just where you're alerted.

What can I do from the brief on mobile?

Open it, click a choice (recorded instantly), leave a note, or archive to signal you've reviewed and decided. The agent watching the brief acts on each event.

Does the agent wait for me?

The agent keeps a monitor running on the brief and picks up your selections and notes as events, so it proceeds the moment you answer — from wherever you are.

Move the gate from your desk to your pocket

Create a free workspace Watch an agent say hello

Or hand this to your own agent right now:

Read https://docs.meltbox.ai/quickstart and https://docs.meltbox.ai/briefs, then walk me through: creating a free Meltbox workspace at meltbox.ai, minting an invite link, redeeming it, and pushing me a brief with a real approve/hold choice about what we're working on — then watch it and act on my decision.

More guides

Get a push when Claude Code needs a decision →The alert that started this. Zero-inbox for agent output →Triage many agents like email. HumanLayer alternatives (2026) →The human-in-the-loop landscape, sourced. Meltbox vs. building it yourself →The DIY approval stack, honestly costed.

Sources

Meltbox briefs, triage keys, MCP card loop: Authoring briefs, Quickstart, MCP server (verified Jul 2026).