Roo Code 3.30.2 Release Notes (2025-11-06)

This patch adds Bedrock Global Inference, expands model options and limits, improves UI clarity, and fixes provider selection and auditing issues.

QOL Improvements

File path tooltips show full paths on hover across reads/edits, directory listings, and search, reducing confusion in narrow layouts (thanks da2ce7!) (#9030)

Eliminates brief UI flicker when cancelling a task; the task view remains stable (#9037)
Pinned models remain visible at the top while scrolling long lists for quicker access (thanks XiaoYingYo!) (#8813)
Checkpoints always commit to Roo’s shadow repository even when GIT_DIR is set in Dev Containers, preventing leaks to external repos (thanks heyseth, nonsleepr!) (#8811)
Restores correct 32K maxTokens for Claude Opus 4.1 to avoid premature truncation on long generations (thanks kaveh-deriv!) (#9046)
Fixes dynamic provider model validation so switching providers uses a valid default model and avoids TypeErrors (thanks NotADev137!) (#9054)
AWS Bedrock requests now report full user agent details in CloudTrail for better auditing and troubleshooting (thanks ajjuaire!) (#9043)

Optional pre-push full test run via RUN_TESTS_ON_PUSH=true in .env.local; default behavior keeps pushes fast (tests skipped) (#9055)

Sets Claude Sonnet 4.5 as the default where available across Anthropic, Bedrock, OpenRouter, Claude Code, and Unbound (#8922)
Updates Cerebras zai‑glm‑4.6 limits to 40,960 output tokens with a 131,072‑token context window; reflects faster response rates (~1,000 tokens/s) (thanks sebastiand-cerebras!) (#9024)
Adds Qwen3-Embedding-0.6B (1024‑dim) and Qwen3-Embedding-4B (2560‑dim) to OpenRouter for code indexing (thanks dmarkey!) (#9060)