[FEATURE] Render cross-session (SendMessage) messages expanded by default, or add a setting

Status Fixed / completed
Reported on v2.1.224
Maintainer reply None cached
Activity 1 comment · opened Aug 7, 2026 · closed Aug 12, 2026

Preflight Checklist

  • [x] I have searched existing requests and this feature hasn't been requested yet
  • [x] This is a single feature request (not multiple features)

Problem Statement

An incoming cross-session message (the native SendMessage/ListAgents lane, v2.1.224, Linux TUI) renders in the recipient session as a single collapsed line:

› Message from @peer (ctrl+o to expand)

The full text is hidden until the human presses ctrl+o, and there appears to be no setting to change this.

Why this default hurts:

  • The human misses the message entirely. In supervision/orchestration workflows (one session overseeing workers, peers coordinating), the person watching a session sees one dim line while the model reads the full text and acts on it. The human becomes the least-informed party about what just entered their session's context — inverted from every other message type, which renders in full.
  • One line hides both sender context and content. Whether the peer asked something, reported something, or requested an action is invisible without a keypress on a window the human may not even have focused.
  • Asymmetry with what the model sees. The model receives the complete <cross-session-message> block. When a human later asks "why did you do X", the cause can be a message they never saw expanded.

Repro: two interactive sessions on one machine, one idle; SendMessage from the other to the idle peer (discoverable via ListAgents); observe the recipient renders only the collapsed line while the model acts on the full text.

Proposed Solution

Render incoming cross-session messages expanded by default, or add a setting (e.g. crossSessionRenderExpanded: true, next to crossSessionInbound) so users can opt into full rendering.

Alternative Solutions

Keep collapse as an explicit opt-in for high-traffic multi-agent setups. Any of these resolves the core problem — hidden-by-default is the wrong default for a channel that injects instructions into a session's context.

Use Case Example

We run multi-agent supervision on top of Claude Code (an orchestrator session messaging worker sessions, peers coordinating across terminals). Since 2.1.224 we prefer the native lane over terminal keystroke injection precisely because it doesn't take over the recipient's input — but the collapsed render means the human operator watching a worker window doesn't see what a peer just told it without an extra keypress per message.

Additional Context

Observed on 2.1.224, Linux, terminal TUI. Related-but-different rendering issues: #73000, #67071 (other content collapsed/not rendered; neither covers cross-session messages).

View original on GitHub ↗

This issue has 1 comment on GitHub. Read the full discussion on GitHub ↗