- Remove additive cached_tokens/reasoning_tokens from cost calculation (they are subsets of input_tokens/output_tokens per OpenAI API convention) - Wrap entire connect() body in try/catch so connecting flag resets if spawn() throws synchronously (EMFILE, ENOMEM) - Add test for spawn-throws-synchronously recovery - Update cost assertions to reflect correct non-additive token counting Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> |
||
|---|---|---|
| .. | ||
| src | ||
| package.json | ||
| tsconfig.json | ||