By name: flash-gv2 description: Decides when to flash GV2, how to verify the model, and when to escalate. Use when flashing GV2, verify model flashed, or @flash-gv2.
Flash GV2 (judgment)
Run steps: @flash-gv2 command. Operator SoT: README.md flash section (paths, hex, xmodem log). Do not duplicate bash blocks here.
Read notes/_hardware_vst.md via .cursor/skills/read-notes/SKILL.md before stating pin or SRAM facts.
When to flash
- New
int8_vela.tfliteafter Colab train/quant (.cursor/skills/train-colab-vela/SKILL.md) - Firmware rebuild with new
APP_TYPEor model path ingv2_firmware/ - Not for UART-only debug (
.cursor/skills/uart-debug/SKILL.md)
Verify (flash succeeded vs model wrong)
| Signal | Meaning |
|---|---|
xmodem 100% on expected .tflite |
Image and model slot written |
| No INVOKE on serial | Wrong firmware build, wrong model path, or UART issue: escalate @uart-debug |
| INVOKE but bad classes | @model-test or retrain; flash may be fine |
After success
Offer: run @model-test on a known image. If user logs milestones, suggest 1 to 2 bullets for notes/_timeline_vst.md (user writes in Obsidian).
Git
Submodule changes: @security-audit then @push-submodule. User provides exact commit message.
Do not
- Invent flash offsets or model filenames
- Push to upstream Himax