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

Sora AI Video Economics: $20 User Costs OpenAI $65 in Compute
OpenAI's Sora AI video generation app reportedly costs $65 in compute per $20/month user, with peak inference costs estimated at $15 million daily versus $2.1 million total lifetime revenue.

User Reports Sonnet 4.6 Outperforms Opus 4.6 for Practical Coding Tasks
A developer testing Claude AI models found that Opus 4.6 produced over-engineered solutions with performance gaps, while Sonnet 4.6 delivered more careful, efficient fixes with lower token usage.

OpenClaw 2026.4.29 Breaks Setups: CPU Spikes, Tool Restrictions, and Fixes
OpenClaw 2026.4.29 introduces CPU spikes from active-run steering, restricted tool profiles breaking exec/fs commands, and stricter group chat handling. Rollback or apply targeted fixes.

Apple Builds New AI Architecture on Google Gemini Foundation Models
Apple announced a major overhaul of Apple Intelligence, built on foundation models co-developed with Google using Gemini technology. The new architecture includes an orchestrator, on-device and server-side models, and multimodal capabilities.