Claude Code v2.1.211 ships small but useful agent workflow tweaks
Claude Code v2.1.211 has been released. The standout change is a new --forward-subagent-text flag (and matching CLAUDE_CODE_FORWARD_SUBAGENT_TEXT ), which lets subagent text and thinking be included in stream-json output.
This makes debugging easier for CI, SDK, and -focused workflows. Other changes include: explicit model overrides now persist when subagents are resumed or given follow-up tasks; MCP servers connect more reliably after idle web sessions wake back up; "always allow" permission rules are now saved at the repository root so they carry over across ; and additional fixes were made for , auto mode hooks, , provider-specific , and on Bedrock, Vertex, Mantle, and Foundry.
Key points
- New --forward-subagent-text flag (plus env var) surfaces subagent text and thinking in stream-json output
- Explicit model overrides persist across subagent resume/follow-up
- MCP server reliability improved after idle web sessions wake up
- "Always allow" permission rules now saved at repo root, shared across
- Additional fixes for , auto mode hooks, , and