Scott Werner 5983c52434 test(llm): cover non-streaming cached-token parsing in openai-compatible ... Mirror the existing stream_chunk_usage_parses_cached_tokens test for the non-streaming ApiResponse path. The non-streaming complete() flow uses a different deserialization target (ApiResponse vs StreamChunk) and a different TokenCounts construction site, so a regression in either is now caught by its own test rather than only by the live e2e test. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>		2026-05-28 16:53:34 -04:00
..
crates	test(llm): cover non-streaming cached-token parsing in openai-compatible	2026-05-28 16:53:34 -04:00
packages/fabro-api-client	refactor: Remove inbound IP allowlisting (#443)	2026-05-27 22:29:08 -04:00