You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Show custom subagent type, usage and context in Agents, plus an at-a-glance status line for orchestrator threads
#33485
Thanks for building T3 Code, I'm really enjoying it. I run long Claude Code threads where one orchestrator launches custom subagents (a planner, a worker and a reviewer) in the background, sometimes for hours. If I've missed a setting that already does any of this, please point me to it.
The problem
While a run is going, I can't easily tell which agent is running or how much room the parent has left. Subagent rows show the model and effort, such as "opus-hi", so my custom reviewer and worker look identical. I filed the wrong effort label as a separate bug.
Would be cool:
1. Show the subagent's type. Show the custom agent's name, such as reviewer or worker, on its row and in the Agents panel, next to its task and elapsed time.
2. Show tokens and context per subagent. Show tokens used and, if possible, how full
that subagent's context is, so I can spot one that's near its limit.
3. An optional always-visible status line for the parent thread. One line with model,
effort, context %, estimated cost, and 5-hour and weekly limit %. I know the context
indicator and Usage page exist. What I'm after is seeing it all together at a glance, so I
know when to start a fresh thread.
Something like this (numbers made up):
Opus 5.5 · high · ctx 42% · $18.20 · 5h 31% · 7d 64%
reviewer · Opus 5.5 xhigh · Review the plan · 6m · 38k tokens
worker · Opus 5.5 high · Build the change · 12m · 91k tokens
## The data appears to exist already
I get most of this today from a small Claude Code status-line script.
Claude Code passes these documented fields to the script:
- `model.display_name`, `effort.level`
- `context_window.used_percentage`
- `cost.total_cost_usd`
- `rate_limits.five_hour.used_percentage`, `rate_limits.seven_day.used_percentage`
For subagents, under `<session>/subagents/`:
- `agent-*.meta.json` has `agentType` and `description`
- `agent-*.jsonl` has a `usage` block on each assistant message, plus the effort the
subagent actually ran at
That subagent file layout is undocumented and could change between Claude Code versions,
and I realize T3 Code may not receive the status-line fields the same way. I'm mainly
pointing out that the information exists and how useful it is.
## Related ideas
"Proposal: turn the thread details card into a thread dashboard" and "Show token usage and
estimated cost per thread" overlap with item 3. Items 1 and 2 are the part I didn't see
requested yet.
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
Uh oh!
There was an error while loading. Please reload this page.
Thanks for building T3 Code, I'm really enjoying it. I run long Claude Code threads where one orchestrator launches custom subagents (a planner, a worker and a reviewer) in the background, sometimes for hours. If I've missed a setting that already does any of this, please point me to it.
The problem
While a run is going, I can't easily tell which agent is running or how much room the parent has left. Subagent rows show the model and effort, such as "opus-hi", so my custom reviewer and worker look identical. I filed the wrong effort label as a separate bug.
Would be cool:
1. Show the subagent's type. Show the custom agent's name, such as
reviewerorworker, on its row and in the Agents panel, next to its task and elapsed time.2. Show tokens and context per subagent. Show tokens used and, if possible, how full
that subagent's context is, so I can spot one that's near its limit.
3. An optional always-visible status line for the parent thread. One line with model,
effort, context %, estimated cost, and 5-hour and weekly limit %. I know the context
indicator and Usage page exist. What I'm after is seeing it all together at a glance, so I
know when to start a fresh thread.
Something like this (numbers made up):
All reactions