# Handled input reproduction

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

Real web client, Chromium, 1280 x 800 CSS pixels, light theme, real Pi CLI 1.5.2.
Each app uses its own fresh database with T3CODE_DEV_SEED=0; no live data copied.
Base: 977d613611be70026b3d8c960254268f5f71ac3d
Head: 5c727d1b659192134252db5977ecae202b1c5c96

Create the project with seed.py. Add the same project to both apps.
Enable Pi and set launch arguments to --approve --no-mcp so the project extension loads.
Select the same available Pi model on both sides. The input is consumed locally.

The browser automation performed these identical actions on each side:

1. Open a fresh new-thread draft.
2. Open the project combobox in the heading and select Handled input.
3. Close the thread-details panel.
4. Type hello into [role=textbox][contenteditable=true].
5. Start browser recording.
6. Click button[aria-label='Submit message'].
7. Observe the turn without sending or stopping anything.
8. Stop recording. The published recordings show the first 12.95 seconds.

Before: Working/Thinking persists; the stop control remains available and no assistant activity appears.
After: startup ends, Working/Thinking clears and the send control returns with no assistant activity.

Gray rectangles redact the model picker in its initial and final positions. They do not cover status,
message content, or the submit/stop controls. Automatic title generation chose different titles.

Tools: real dev server, built-in browser interaction and recording tools, ffmpeg for video redaction,
visual frame inspection, and ghdrop upload. No tests are offered as evidence.