Claude Code 2.1.323 drops a setting that showed what each subagent costs
A GitHub issue says Claude Code's latest release removed a documented setting for monitoring background agents, with no warning, reason or replacement.
The short version
- Claude Code 2.1.323 removed the subagentStatusLine setting; 2.1.322 still runs it.
- The changelog notes the removal, but nothing warns at startup. Configured commands just stop running.
- The setting showed per-agent context use, spend, model, reasoning effort and cache state.
- The issue says the docs still describe the setting as live, and a request for a replacement hook has gone unanswered.
- Anthropic's reasoning is not in the issue. 'Careless' is the reporter's word.
Why it matters
If you run several background agents in Claude Code, this setting was the one place to see which agent was spending money or running out of room. The episode is a reminder that anything you build on one feature of a fast-moving tool can break in a routine update. It also shows why it is worth reading release notes before you upgrade.
What the issue says happened
As github.com reported through a user-filed issue on Claude Code's repository, version 2.1.323 removed a setting called subagentStatusLine. A subagent is a helper that Claude Code starts to handle a side task in the background. The setting let users attach their own command that drew a custom status row for each of those helpers.
The changelog for 2.1.323 does say the setting was removed. It adds that agent rows now always use the default look, and that settings files still containing the key load as before. The settings schema, the file that describes each option, now labels the key "Deprecated: no longer has any effect."
The complaint is about how it was done. According to the issue, nothing warns you at startup. A command you configured simply stops being called, and you notice when your agent panel goes blank. Version 2.1.322 still runs the command; 2.1.323 never does.
What users lost
The issue argues this was the only place in Claude Code where you could see, for each running subagent, a set of details at a glance. These were:
- how much of its own context window it had used (the amount of text the model can hold in mind at once)
- what it had spent so far, in tokens and dollars
- which model it was using and at what reasoning effort
- its status, its age and its prompt-cache state (whether a stored copy of earlier text is still saving money)
The issue shows a sample row from 2.1.322. It reads 8% context use, 82.6K of 1.0M, a running total of $2.06, a model labelled opus at medium effort, five minutes forty seconds of run time and one error. After the update, the same agent shows only its name and a line like "Running grep."
The author's point is that with several agents running, the default row does not tell you which one is burning context, which is near its limit, which one switched to the expensive model, or which one is paying again for an expired cache.
Why the reporter calls it a contradiction
The issue notes that the changelog kept adding to this setting right up to its removal. Earlier entries added reasoning effort to the data the command receives, added an agent type so scripts could tell custom subagents apart, and fixed a bug with plugin-provided commands. People built on those additions, the author says.
The author also says the official docs pages for the status line and for settings still present the setting as live. We have not checked those pages ourselves, so treat that as the reporter's account.
Another comment on the issue, from a different user, says they had built real-time observability on the feature and call the removal unacceptable. That is a strong reaction, but it is one person's.
Why mods do not fill the gap
Claude Code has a mod system for customizing the interface. The issue lists the places where a mod can draw things in version 2.1.323, such as user messages, tool output, the spinner and the area above the prompt. None of them is a row in the agent panel, so the author says no mod can recreate what this setting did.
A request for exactly that drawing spot, numbered #105919, was filed on 2026-10-29 and, per the issue, has had no answer. Four other open requests (#102183, #102890, #96895 and #97634) assume the setting exists. The last notes that no other place in the interface shows a subagent's model.
What the reporter wants, and what is unknown
The issue asks for three things. First, restore the setting in the next release with the same data as 2.1.322. Second, if the agent panel is being redesigned, ship the replacement before removing the old hook. Third, never remove a documented, configurable feature without a release that warns at startup and a changelog line that explains why.
What we do not have is Anthropic's side. The changelog entry states the removal but gives no reason, and the issue includes no reply from the company. Words like "silently" and "carelessly" are the reporter's framing. The facts that hold up are the removal itself, the changelog wording and the version numbers. Whether it was a deliberate redesign step or a misstep is still open.
What to watch
- Does Anthropic respond to the issue, restore the setting or explain the removal?
- Will a replacement drawing spot for agent-panel rows (request #105919) ship?
- Will the docs be updated to match the removal?
Try it
Run Claude Code's version check. If you use a custom subagent status command and your agent rows went blank after updating, 2.1.323 is the likely cause. Before future updates, skim the changelog for the word 'Removed'.
Sources and links
- Reportinggithub.com: the article we sourced
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.
Listen to the AI Wire headlines
Listen to the AI WireListen to our daily AI shows every morning: on your commute, in the shower, wherever you are
Listen to the daily showsFirst reported by github.com · Read the original. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
