fix(orchestrator): Skip blank subagent opening messages

[mwolson]

Carried from another PR

• fix(orchestration): Preserve Claude subagent attribution after settle —
  carried for #5388 claude-postsettle-attribution.
  Review it there, not here.
• fix(orchestrator): Skip blank subagent opening messages — this PR.

Summary

• Stop provider-native subagent child threads from opening with an empty or
  whitespace-only user message under the "Sent by another agent" label.
• Gate every subagent opening message with a shared trim-aware check, and emit
  Claude's opening message the first time the task prompt has real text rather
  than on first registration.
• One unique commit above fix(orchestration): Preserve Claude subagent attribution after settle #5388 claude-postsettle-attribution on
  t3code/codex-turn-mapping.

Problem and Fix

Problem and Why it Happened	Fix
A Claude Agent launch whose prompt is whitespace-only still produced a child user message. The Agent tool accepts it, task_started carries it, and the emission site had no emptiness guard, so the web timeline rendered an empty bubble.	Shared hasSubagentPromptText trims before deciding. Every makeSubagentConversationArtifacts caller with role: "user" goes through it (Claude, Codex, Cursor, ACP).
A task_progress frame can register a Claude subagent before any prompt-bearing frame. Registration fell back to prompt: input.prompt ?? "", and the opening message was tied to first registration, so it was emitted as "" and never rewritten when a later task_started supplied the real prompt.	Claude emits the opening message when hasSubagentPromptText(task.prompt) && !hasSubagentPromptText(priorTask?.prompt), so a late prompt still opens the child exactly once and a prompt change from one non-empty value to another does not re-emit.
Codex only checked input.prompt.length > 0, which still admits whitespace. Cursor and ACP had no guard.	Codex's length check is replaced by the shared predicate; Cursor and ACP pick up the same gate.

Validation

• New replay fixture claude_subagent_empty_prompt covers both routes (whitespace Agent prompt and task_progress before task_started). Confirmed fail before the emission fix and pass after.
• Replay suites (OrchestratorReplayFixtures.integration / .contract, ClaudeReplayFixtures.integration, CodexReplayFixtures.integration): 89 tests, 1 skipped.
• Adapter and projection suites (Claude, Claude testkit, Codex, Cursor, Cursor testkit, ACP, ACP registry, Grok, SubagentProjection): 274 tests.
• vp check: passed.
• vp run typecheck: passed all 15 packages.
• Isolated private serve live check against this branch: a whitespace Agent launch produced a completed subagent row with prompt: " " and a child thread with no user message; a sibling ordinary-prompt child held exactly one :prompt message. Zero trim-empty user-role projection messages in the isolated DB.

Checklist

• This PR is small and focused
• I explained what changed and why
• I included before/after screenshots for any UI changes (not applicable: server-side projection only)
• I included a video for animation/interaction changes (not applicable)

---

Note

Low Risk
Server-side projection-only change with a narrow guard; low risk to auth, data, or core runtime paths.

Overview
Subagent child threads no longer get an opening user message when the task prompt is empty or whitespace-only, avoiding empty “Sent by another agent” bubbles in the timeline.

A shared hasSubagentPromptText (trim-aware) gates makeSubagentConversationArtifacts for user-role opening messages across provider adapters. In this diff, ACP only emits message.updated / turn_item.updated for the :prompt item when update.prompt has real text after trim.

Related adapter behavior (same PR): Claude defers the opening message until the first frame with non-empty prompt (not first registration); Codex drops a raw length check in favor of the shared helper; Cursor and ACP gain the same guard.

^{Reviewed by Cursor Bugbot for commit 0e1b4fe. Bugbot is set up for automated code reviews on this repo. Configure here.}

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 780c5d1e-0f63-4320-a3ce-9608ba6976d1

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}