Playbooks · Coding tasks
OpenCode 2.3.0: subagents move to the background
Written November 13, 2026
In 2.3.0, subagents run quietly in the background while you keep working, and transcripts fold into tidy activity rows.
What changed
OpenCode 2.3.0, released November 13, changes the default shape of a session. Subagents stop blocking the parent session, and the transcript stops showing every step by default. Both changes are about the same problem: a session that makes you watch work you delegated, in full, is a session you cannot use for anything else at the same time.
Background subagents by default
The headline in the 2.3.0 notes is "Background Subagents by Default". Subagents now return immediately, continue in the background, and report back whenever they become idle. You can keep working in the parent session while multiple subagents run concurrently. In the core list, the same item is stated more precisely: subagents always run in the background, return their session ID immediately, and report results to their parent when they become idle.
One fix protects the new default from its obvious failure mode. When a subagent goes idle while its own background work is still running, its report to the parent now says it is still running, so the parent waits for the final report instead of treating the partial answer as done. A background model is only useful if "idle for a moment" and "finished" stay distinct, and this release keeps them distinct.
Quieter transcripts
The second headline is "Quieter Transcripts by Default". The terminal interface now summarizes each run of tool calls and thoughts into one collapsible activity row, so the answer is easier to follow. The detail is folded, not deleted: to see every step again, set Verbosity to medium or high in settings, or set session.verbosity in your TUI config. An existing setting is kept, so anyone who already chose a verbosity level will not be moved by the update. The practical trade is clear. The default view now answers "what did it conclude" first, and the step-by-step record is one setting away when a result looks wrong.
Smaller items in the same release
- In the desktop app, Jump to message, from the command palette or Cmd+Alt+G (Ctrl+Alt+G on Windows and Linux), searches prompts across the current session and jumps straight to one.
- Shell commands now survive server handoffs and keep their output when interrupted. Tool output truncation keeps the beginning of the output and preserves a usable preview when the full output cannot be saved.
- Two items sit behind flags and are named experimental in the notes: support for MCP events, behind the
mcp_eventsflag, and moving a shell command to the background when its timeout expires, behind theshell_timeout_backgroundflag. With that flag on, an explicit timeout on a background command stops it. - Providers can present a client certificate for gateways that require mutual TLS, set through
settings.tlson the provider. Local model discovery now waits up to 15 seconds per request instead of 1, so remote or tunnelled endpoints list their models. - One Copilot fix lands in this release: GitHub Copilot requests no longer use the wrong endpoint before models finish loading. There is no Copilot Auto item in the 2.3.0 notes; that change belongs to the earlier 2.2.7 line.
What to do
Update and try a delegated task you would previously have run in the foreground; the parent session stays yours while it runs. If the folded transcript hides more than you like, raise Verbosity rather than looking for the old layout. Leave the two flagged experiments off unless you are testing them on purpose.
Related playbooks
- Move OpenCode 1.x to 2.0 without losing config
- Have one agent write the code and a second agent check it in the background
- Stop fork subagents inheriting your context
- Measure what Claude Code's background-agent summaries cost you, and decide what to turn off
Sources
Back to: Coding playbooks hub · Also: OpenCode 2.2.7: Copilot Auto and breaking changes · Move OpenCode 1.x to 2.0 without losing config