Some gateways report all-zero usage in stream-json while the Claude
transcript carries the real sample ~100 ms later. Delay those assistant
messages briefly and patch usage from the main transcript so context and
usage meters stay truthful.
via [HAPI](https://hapi.run)
Co-Authored-By: HAPI <noreply@hapi.run>
Co-Authored-By: Claude Code <noreply@anthropic.com>