loongson/pypi: links for agent-trainer