one more

one more time

Copied source project

Add deployment marker to verify Space is running latest code

Debug: log final config string sent to grader before evaluation

Debug: log raw model output to inspect formatting issues

Optimize LLM prompts and model selection for better config-debug task solving

Fix: add guaranteed proxy validation call immediately after client creation

Fix: use strict os.environ[] for API_KEY and API_BASE_URL to match validator requirements

Fix: use gpt-4o-mini as default model for better proxy compatibility

Replace repo with updated config-debug files