Claude Code v2.1.211: Subagent Text Forwarding, Permission Fixes, and Vim Mode Changes

Claude Code v2.1.211 is out with a mix of new features, security fixes, and quality-of-life improvements. The headline addition is the --forward-subagent-text flag (and corresponding CLAUDE_CODE_FORWARD_SUBAGENT_TEXT environment variable) that includes subagent text and thinking in stream-json output — useful for headless or scripted workflows that need visibility into subagent reasoning.
Key Changes
- Subagent text forwarding: New flag
--forward-subagent-text/ env varCLAUDE_CODE_FORWARD_SUBAGENT_TEXTincludes subagent text and thinking in stream-json output. - Security fix for permission previews: Permission previews relayed to chat channels now neutralize bidirectional-override, zero-width, and look-alike quote characters, preventing tool inputs from visually altering the approval message.
- PreToolUse hook 'ask' decision honored: Auto mode no longer overrides a PreToolUse hook's
askdecision for unsandboxed Bash. A hookasknow floors the decision at a prompt. - Subagent model override persistence: Subagents spawned with an explicit model override no longer revert to the parent's model when resumed or sent a follow-up message.
- Vim mode improvements:
sandS(substitute char/line) now work in NORMAL mode, matching vim behavior. - Session cost reset:
/clearnow resets the session cost counter — the statusline's cost starts at $0 after/clear. - Integer env vars: Environment variables for timeouts, token budgets, and retry counts now accept scientific notation and digit separators like
1e6and64_000. - Background agent reporting: Improved — Claude now reports the status of still-running agents and waits for real completion instead of fabricating results.
- Prompt caching regression fixed: Fixed a caching regression on Bedrock, Vertex, Mantle, and Foundry that billed the trailing system context block as fresh input tokens on every request.
- File upload validation: Filenames ending in DOS device suffix (
.prn) or trailing dot are now accepted; files with multiple hard links are refused. - Always allow rules: Permission rules now save at the repository root, so approvals granted in a git worktree persist across sessions and worktrees.
This release also fixes a number of edge cases: parallel sessions logging out after wake-from-sleep, plugin MCP servers not reconnecting after idle web sessions, subagent model overrides reverting, nested .claude/rules/*.md files loading when excluded, and more.
For VSCode users, the Remote Control banner has been updated to clarify what it does. Claude in Chrome gets hardened file-upload path validation and a working save_to_disk on screenshot actions (now writes the image to disk and returns the path).
If you're using Claude Code on Vertex or Bedrock, note the fix for a spurious fallback notice when a model is explicitly configured — the client no longer attempts the default Opus model at startup.
📖 Read the full source: GitHub Claude-Code
👀 See Also

Control-UI LAN Access Issues in Docker OpenClaw Bridge Networks
A user reports persistent problems accessing OpenClaw's Control-UI via LAN connections in Docker bridge networks, with version 2026.3.14 briefly supporting token-based access before subsequent versions reverted to requiring pairing and throwing scope errors.

Apple Using Google Gemini Access for On-Device AI Model Distillation
Apple has full access to Google's Gemini model for distillation, creating smaller on-device AI models for Siri and other features in iOS 27 without internet connectivity.

Automating Social Media with OpenClaw: Possibilities and Discussions
A discussion on Reddit explores the potential of automating social media tasks using OpenClaw.

Claude Code 2.1.63 adds bundled slash commands, HTTP hooks, and memory leak fixes
Anthropic released Claude Code 2.1.63 with 26 CLI changes including new /simplify and /batch slash commands, HTTP hooks that POST JSON to URLs, and fixes for multiple memory leaks in long-running sessions.