# Repeated linked-worktree import proof

Use the scenario runner from current main. Apply scenario-startup.patch to its start-app.ts: it enables Codex transcript discovery while retaining the no-model executable, so no Codex turn can run. Copy repeated-worktree-import.e2e.ts into scripts/scenarios/.

Run after vp i:

```sh
CODEX_HOME=/tmp/factory-pr-36608-evidence-2/codex \
T3CODE_DEV_SEED=0 \
T3_SCENARIO_INHERIT_ENV=CODEX_HOME,T3CODE_DEV_SEED \
vp run scenarios:prove scripts/scenarios/repeated-worktree-import.e2e.ts \
  --base 28f178682949e0eaf732a6a8314cb7c2ba94dcc9 \
  --head b0c2009f333433f7fe7340d80f23b1a7d3795109 \
  --upload-command 'ghdrop upload'
```

The scenario creates synthetic native Codex transcripts, not database rows. It uses 101 distinct session identities and timestamped, readable user/assistant pairs in a linked Git worktree. The app starts with a fresh database on each commit. The scenario imports twice through Settings → About → Welcome page → Open → Review all → Select all → Import, opens conversation 001 on both sides, searches for 101 after the second import, and opens conversation 101 on the head. The right panel's linked-history branch belongs to the linked worktree.

Environment: Linux, Chromium, web client, 1440×1000, dark theme, reduced motion. The runner independently installs and starts each app revision. Captures wait for two identical screenshots at least 500 ms apart; result toasts are allowed to leave. The expected no-model provider warning is dismissed using its UI button.
