v0.9.2

Fixes

• Streaming tool_use: forward delta.tool_calls through SSE stream (backend)
• DeepSeek-R1 reasoning: forward reasoning_content as thinking blocks (backend)
• Login page bug: wrap backend errors in Anthropic format so Claude Code shows the error instead of a login prompt
• max_tokens: always inject default (4096, adaptive) to prevent 400 on Turn 2+
• Token stats: search full SSE response for output_tokens, not just last chunk
• Streaming memory: cap accumulated response at 5MB
• Version: read from package.json at runtime
• Port validation: reject invalid ports (1-65535)
• Error messages: better errors for balance and models commands
• Model pricing: complete pricing table for all router-referenced models

Upgrade

npm update -g @blockrun/cc