Your subagents,
as a tiny pixel crew.

A Claude Code mod that shows who is doing what, how far along they are, and how long is left. Right above your prompt.

/plugin install agent-crew --marketplace bekir1184/agent-crew
Costs nothing: it never calls a model and never touches your prompts.
The crew at work: agents search, read, write and think, their bars fill up, and each raises a checkered flag when done
Claude, live

See what Claude is doing, right now

While Claude works, a framed line shows its current step: searching, reading, writing, running tests.

When the turn ends it shrinks to one quiet line with the totals. The arrow opens what happened.

The AGENT line moves through each step, then becomes a thin line with the turn's totals
The crew

Every subagent gets a critter

They line up under Claude, one row each, and hold the tool they're using. Helpers tuck in under their parent.

Searchinga magnifier
Readinga page
Writinga code window
Runninga laptop
Browsinga globe
Thinkinghourglass, cloud, bulb
Startinga hard hat
Asking youa question bubble
Donethe checkered flag
Faileddizzy eyes
Cancelledfast asleep
└
Helperstucked under their parent
Thinking...

Never just a spinner

Agents spend most of their time thinking, so the critters keep it lively: an hourglass, a thought cloud that fills with a question mark, a lightbulb that lights up.

Each one at its own pace, so the crew is never in lockstep.

Three critters thinking in different ways
It needs you

Waiting on your OK? You'll see it

When an agent stops to ask for permission, it turns yellow with a question bubble, and the title says how many are waiting on you.

A row turns yellow: Needs approval
The details

One arrow, the whole story

Each row has an arrow. It opens everything the row had no room for: the full task, its to-do list, the files it touched, and its tokens.

A row opens into a details panel
How long is left?

A good guess, for free

Agent Crew remembers how long this kind of agent took before, and counts down from there. Running long? It re-estimates instead of going negative.

Past run times, the middle one picked, a countdown and a re-estimate
0past runs remembered, per agent type and model
0tokens spent on guessing: it's plain arithmetic
0second between updates, without ever slowing Claude down
Honest tokens

Only the work counts

A row shows what the agent's own work took: what it read in and what it wrote.

Claude also keeps the conversation in a cache and re-reads it on every request. That's not the agent's work, so it lives in the details panel instead of inflating the row.

the agent's workcache re-reads
EXPLORE  Map the auth flow3.1k tok

Put a crew above your prompt

/plugin install agent-crew --marketplace bekir1184/agent-crew
Then type /crew demo for a 30-second show.