# Foreground subagent reproduction

AI-assisted; verified with the tests and tools listed above.

Use two isolated checkouts: base a4d0a06e7f1a062b2fcbc93e5052db017483297e and head f33697872f759c27ab53f0a70978aae0cde405c9.
Install each with `vp i`. Install `@opencode/cli@2.2.7` in a scratch directory.
Set OPENCODE_BINARY to its opencode2 executable and MODEL to the same authenticated inexpensive model on both sides (runtime identifiers intentionally omitted).
Run setup.py in each checkout, then `T3CODE_DEV_SEED=0 vp run dev`.
Create an empty synthetic Git repository and set WORKSPACE to it. Register it on each server:
`node apps/server/src/bin.ts project add "$WORKSPACE" --title 'Foreground subagents' --base-dir .t3`.
Run reproduce.sh from each checkout. It sends the identical prompt through the real server CLI.
Open each startup pairing URL in the real browser, then open the returned thread.
Use the same 1280 x 800 CSS viewport, light theme, sidebar open and thread details open.
During the three-child wait, record about 23 seconds with the subagent group collapsed and Lineage visible. No clicks during either recorded wait.

For public evidence, hide the model selector with `button[aria-label*="Claude"]{visibility:hidden!important}` and hide the leaf showing the environment's machine label. Neither redaction touches timeline, status, child list, or Stop.
Videos are the timing evidence. PNGs are poster frames extracted from those videos, not claims of an animation-free still.

Observed base: 3 subagents / 3 working, Lineage 3 running, live Thinking label, Stop generation.
Observed head: 3 subagents / 3 working, Lineage 3 running, live Waiting on 3 subagents label, Stop generation.
Read-only SQL during each capture: parent run status running; three subagents with origin provider_native and status running.
Head subsequently completed with done-1, done-2, done-3.

This is a timeline label fix. Sidebar Working and busy composer remain on head. Queued-follow-up delivery was not exercised.
