Claude Code 2.1.237 restores prompt caching through LLM gateways
The fix ends silent full-price billing for anyone routing Claude Code through a proxy — plus a new Concise output style, as OneCLI opens a team agent harness.
Copy markdown
Caching survives the gateway hop
2.1.237 fixes prompt caching for sessions running through an LLM gateway or custom base URL. If you route Claude Code via LiteLLM, a Bedrock proxy, or OpenRouter, cached tokens had been silently billing at full price — update and stop overpaying on every turn.
A Concise mode that skips the preamble
The same release adds a built-in 'Concise' output style: Claude leads with the result and drops the running narration while still working thoroughly. Fewer wasted tokens per turn, and less to scroll past on long runs.
Elsewhere: OneCLI open-sources a team agent harness
A YC S26 Launch HN, Apache-2.0: every teammate gets a sandboxed agent whose real credentials are injected at a gateway per request, so a prompt injection can't exfiltrate your keys. Worth a clone this weekend if you're wiring agents into GitHub, Gmail, or Notion.