Done

Messaging-mode team runs keep every finished step's console alive; a 25-step run exhausted Mac memory

From BiTzaDone 1 week ago

In a team run in messaging mode (one console per role), the console of a step that has completed stays alive until the run is finalized or cancelled. Each console is a Claude Code process with its own MCP servers (about 10 node processes each, with the user's CLI configuration inherited). During a 25-step run on a 128 GB Mac, alongside other projects' sessions, macOS ran out of memory: a JetsamEvent killed WindowServer (the whole GUI session). At that moment: 22 Claude Code processes (9.2 GB) and 215 node processes (18.6 GB).

Expected: an option to close (or hibernate) a step's console once it has handed off and no loop-back edge targets it, and/or a cap on live consoles per run.

Actual: every role console stays alive until the run ends.

Steps: create a team with about 20 agent nodes in messaging mode, run it on a ticket, watch the process count grow as steps complete.

AgentsRoom diagnostics

App version: 1.203.0
OS: darwin 27.0.0 (arm64)
Agent CLI: claude
App language: system

Problem to solve

Long team runs in messaging mode accumulate live consoles (each with its own MCP servers) until the machine runs out of memory and the macOS GUI session is killed.

Proposed behavior

Close or hibernate a step's console after its handoff when no loop-back edge targets it; optionally a per-run cap on live consoles and a per-team list of MCP servers for step consoles.

Who benefits

Users running multi-step teams on a single machine with several projects open.

Acceptance criteria

A 25-step run stays within a bounded number of live consoles instead of growing to 20+.

Replies

The team1 week ago

Thanks: your idea is implemented and ships in the next release. We built it the way we pictured it, so if the result is not what you had in mind, tell us and we will adjust it.