Repository navigation
v1.3.37
- rewrite create_jinx_stream as true agentic loop using check_llm_command
- remove weak followup classifier LLM call between iterations
- agent loops autonomously: stream → execute jinxes → feed results back → repeat until stop
- max_followups default bumped from 3 to 10