Field noteNov 13, 20265 min read1 source

opencode v2.3.0 makes subagents always run in the background and adds a quieter TUI transcript

In one minute

OpenCode v2.3.0 makes subagents run in the background by default: they return a session ID immediately and report to the parent when idle. It also reduces default transcript detail in the terminal interface. For teams using OpenCode, the release changes how delegated work can run alongside the main session and how progress is reviewed.

What changed: OpenCode v2.3.0 makes subagents run in the background by default and changes the TUI transcript.

What happened

OpenCode’s v2.3.0 release notes, published on November 13, describe a change to every subagent session: subagents now return immediately, continue in the background, and report back whenever they become idle. The parent session receives the child session’s ID at once, allowing work to continue in the parent while multiple subagents run concurrently. The notes say this behavior is now the default, rather than describing it as an experimental option.

The release also addresses a potential ambiguity in nested delegation. If a subagent becomes idle while its own background work is still running, its report to the parent says that work is still running. The parent can therefore wait for a final report rather than treating a partial result as complete. This is a behavioral safeguard for workflows where delegated agents launch or manage further work.

The terminal user interface (TUI) now defaults to low transcript verbosity. Each run of tool calls and thoughts is summarized as one collapsible activity row, making the answer easier to follow. Users can see every step again by setting Verbosity to medium or high, or by configuring session.verbosity; an existing setting is preserved. The desktop app adds a “Jump to message” command, available from the command palette or with Cmd+Alt+G on macOS and Ctrl+Alt+G on Windows and Linux, to search prompts in the current session and jump to one.

Other additions include experimental support for MCP events behind the mcp_events flag and experimental backgrounding of shell commands when a timeout expires behind shell_timeout_background. The release notes specify that, with the latter flag enabled, an explicit timeout on a background command stops it. They also describe improvements to shell commands surviving server handoffs, freeform tool input on compatible models, and LM Studio decision-model support. The release is identified as v2.3.0 and lists 12 commits to dev since the release; it does not give a price or announce a change to access terms.

What the archive adds

New Runtime’s earlier coverage connects OpenCode’s releases with continuing changes to the tool’s agent workflow: v2.2.2 and its /goal loop and v2.2.4’s session-persistent terminals and opt-in artifacts, both dated October 31, 2026. This is a new release and new workflow detail in an ongoing OpenCode thread, not a restatement of those capabilities.

Trend signal

The release supports goal-scoped loops replace manual continuation only weakly: background delegation and clearer reporting help a parent session coordinate work, but the release does not establish an external completion contract. The more direct signal is that a coding agent’s harness is evolving around session management and orchestration, not just model choice; New Runtime does not yet have a matching listed pattern for that narrower observation.

Why now

The release notes do not explain why these changes were timed for this release. They do make the operational timing clear: subagents can now proceed while a parent session continues, and the parent receives an explicit signal when child work remains unfinished.

Why it matters

Background execution can reduce the amount of time a developer spends waiting for one delegated task before starting another. Returning a session ID immediately gives the parent a handle on the child while keeping the main session available for other work. The explicit warning about a subagent’s own unfinished background work is also important: concurrency is useful only if the orchestrator can distinguish an interim report from a completed result.

The quieter transcript addresses a different constraint. Summarizing tool activity can make the final response easier to scan, but it also hides detail by default. Teams that need to audit tool calls or debug a run should know how to restore medium or high verbosity or inspect their existing session setting. The release provides the controls, but does not report measured gains in task speed, completion quality, or review effort. Builders should treat the update as a workflow capability to test against their own delegation and oversight practices, not evidence that parallel agents automatically improve outcomes.

What happens next

  • OpenCode users will increasingly run concurrent subagent work from a parent session (3 months, 70%). Watch for: subsequent release notes or documentation describing changes to parent-child session orchestration.

Limits and uncertainty

The release notes do not quantify how many users or sessions benefit, or report changes in latency, task success, or resource use. They do not specify concurrency limits for subagents. MCP events and timeout-triggered shell backgrounding are explicitly experimental and gated by flags; the notes do not describe a production-readiness timeline for either. The release notes are the project’s own account, and no independent evaluation of the new behavior is cited here.

Skill to build now

Try a small delegated task in OpenCode v2.3.0: start a subagent, note the session ID it returns, and continue a separate task in the parent session. Check whether the child’s report clearly distinguishes a finished result from work still running. For review-heavy tasks, switch transcript verbosity to medium or high and compare whether the extra activity detail helps you verify the result.

Reader takeaway

OpenCode v2.3.0 makes parallel subagent work the default and clarifies when a child session is not finished. The practical change is not simply “more agents”: it is that the parent session can keep moving while retaining a session handle and a clearer signal about completion. Test that handoff, and keep detailed transcripts available when auditability matters.

Evidence / sources

  1. 1github.com/anomalyco/opencode/releases/tag/v2.3.0
All Field Notes

Next reads

Read next

Open the graph
  1. 01topicMcpExplore the mcp topic hub.
  2. 02archiveField NotesOpen the latest editorial analysis.
  3. 03sourcesSourcesSee the original sources this archive cites.
Who read this page?Machine requests, hidden until opened

Loading the privacy-safe route aggregate...

Open the JSON contract