Harden Dream QA vision intake

#32
by ADJCJH - opened
Build Small Hackathon org

Parse Modal MiniCPM-V pipeline responses, localize common visual anchors for Chinese output, and add three-sketch evidence for image plus text dream intake.

Cannot merge
This branch has merge conflicts in the following files:
  • docs/smoke/2026-06-12-modal-vision-sketch-evidence.md
  • dream_customs/pipeline.py
  • tests/test_pipeline.py

Sign up or log in to comment