Segment 1: remove email refs, generalize UI for any LLM input

Pass HF_TOKEN to pipeline for private model access

Trigger rebuild: pull retrained v2 model weights

Revert to gradio 5.9.1 (4.x broken with newer jinja2/huggingface_hub)

Fix: remove theme (jinja2 unhashable bug), add server_name=0.0.0.0

Fix: gradio 4.44 + pin huggingface_hub==0.23.0 to avoid HfFolder error

Fix: disable SSR mode for HF Spaces compatibility

Fix: remove gradio from requirements (SDK manages it via sdk_version)

Fix: move theme to gr.Blocks(), remove from launch() (gradio 5.x)

Fix: use gradio 5.9.1 (HfFolder removed in newer huggingface_hub)

Fix: downgrade sdk_version to 4.44.0, python to 3.11 to unblock Space build

Switch to fine-tuned Builder117/distilbert-prompt-injection

Pause: placeholder while fine-tuned model trains

Fix Gradio 6.0: move theme param from Blocks() to launch()

Fix colorTo: orange -> yellow (HF only allows named colors)

Add prompt injection detector (protectai/deberta-v3-base-prompt-injection-v2)

initial commit