Understand token usage report from Claude Agent SDK

[yshuolu]

I got this claude agent message in ResultMessage's metadata:

usage={'input_tokens': 1792, 'cache_creation_input_tokens': 53078, 'cache_read_input_tokens': 1346549, 'output_tokens': 20027, 'server_tool_use': {'web_search_requests': 0, 'web_fetch_requests': 0}, 'service_tier': 'standard', 'cache_creation': {'ephemeral_1h_input_tokens': 0, 'ephemeral_5m_input_tokens': 53078}

The cache_read_input_tokens is about 1.3M!

Is this something I need to worry about? The cached read input tokens seem coming from nowhere as my project codebase is pretty small, only thousands chars in the folder.

And is there a way to see breakdown of cached tokens? Like what files are loaded.