How-to · running many agents
Stop checking terminals: a zero-inbox for agent output
Five to ten Claude Code agents across different repos turns into an unmanageable mess of terminal tabs and forgotten sessions. The fix isn't another dashboard you update by hand — it's making the agents deliver their decisions to one inbox you clear like email.
Short answer. Have each agent push its decisions as briefs into one inbox instead of leaving them in separate terminal tabs. You triage the inbox with keyboard shortcuts — move, open, choose, note, archive — and each choice flows back to the agent that asked. One place, cleared like email.
The problem
Review is the bottleneck now — and terminal tabs are a bad queue
Writing code stopped being the constraint the moment agents got good at it. What's scarce now is your attention on what they produced: the review queue is the new bottleneck, and a fleet of agents doesn't drown you in work — it drowns you in decisions.
One agent is a conversation. Five is a switchboard. The output is scattered across tabs, each waiting on a different decision, and there's no single view of "what needs me." Sessions crash and get forgotten; you re-read context to remember where each one was. The usual patches — a tmux grid, a hand-updated kanban, a "check the terminals" habit — are all you doing the aggregation the agents should be doing.
The fix
One inbox the agents write to
In Meltbox, an agent talks to you by pushing a brief — a self-contained page with the decision it needs, clickable choices, and a note box. Every agent, every repo, pushes to the same inbox. You clear it like a zero-inbox mail reader, and each answer routes straight back to the agent that asked.
every agent pushes to the same place
mb brief push ./review.html --title "api: merge PR #214 or request changes?" --kind options --project api
mb brief push ./plan.html --title "web: which caching approach?" --kind options --project web
# both land in one inbox; --project tags keep the fleet legible.triage the fleet with the keyboard
j / k move down / up the inbox
Enter open the focused brief
click record a choice — the waiting agent sees it instantly
n leave a note e archive (= reviewed & decided)
u restore z undo / searchArchiving a brief signals the producing agent that you've reviewed and decided; it acts on your selections and notes, keeps working, and follows up with a new brief for the next decision. The inbox trends toward zero because every item is a real decision with a real answer — not a log line. You can also promote a brief into a durable dashboard or document straight from the reader.
A dashboard shows you state; you still have to notice things and go act. An inbox is a work queue with a reply loop: each item is something to decide, and deciding it advances the agent. That's the difference between watching your fleet and clearing it.
Side by side
Ways to keep track of many agents
| Terminal tabs / tmux | Hand-updated kanban | Meltbox Brief Inbox | |
|---|---|---|---|
| One place to see what needs you | No | if you update it | Yes |
| Agents populate it themselves | No | No | Yes |
| Deciding advances the agent | per tab | No | Yes |
| Keyboard-fast triage | varies | No | Yes |
| Survives a crashed session | scrollback only (if the server survives) | stale | brief persists |
Meltbox behavior verified against Quickstart and Core concepts — Briefs, July 2026.
FAQ
Common questions
How do I manage output from multiple Claude Code agents?
Have each agent push its decisions as briefs into one inbox instead of leaving them in separate terminal tabs. You triage the inbox with keyboard shortcuts — move, open, choose, note, archive — and each choice flows back to the agent that asked. One place, cleared like email.
What's a brief?
A brief is a single self-contained HTML page an agent pushes to your inbox — a plan to approve, options to pick between, a draft to review — with clickable choices and a notes box. Your selections and notes become events the producing agent reads and acts on.
How fast is triage?
Keyboard-first: j/k move, Enter opens, clicking a choice records it, n notes, e archives (which signals the agent you've reviewed and decided), z undoes. It's built to clear like a zero-inbox reader.
Do agents keep working while briefs sit in the inbox?
Yes. An agent pushes a brief and keeps a monitor running; when you act, it reads the events and does the next thing, then follows up with a new brief for the next decision.
Clear your agents like email
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 first brief with the decisions you need from me on our project — then watch it for my replies.
More guides
Sources
- Meltbox Brief Inbox, triage keys, promote-to-dashboard: Quickstart, Core concepts (verified Jul 2026).