[Release] anthropics/claude-code v2.1.219: v2.1.219

Claude Code v2.1.219 introduces Claude Opus 5 as the new default model with 1M context window and improved pricing, adds network security controls via strict allowlist settings, and enhances subagent nesting capabilities up to depth 3. The release includes fixes for permission handling in self-hosted runners, improved MCP server error reporting, and UI improvements across model selection, workflow sizing, and accessibility features.

Key Points

• •Claude Opus 5 is now the default Opus model with 1M context window at $10/$50 per Mtok, replacing Opus 4.7 in fast mode
• •New `sandbox.network.strictAllowlist` setting enables denying non-allowlisted hosts for sandboxed commands without user prompts
• •Added `DirectoryAdded` hook that fires when new working directories are registered mid-session via `/add-dir` or SDK `register_repo_root`
• •Subagents can now spawn nested subagents up to depth 3 (previously depth 1), with environment variable override available
• •Improved MCP server error handling with HTTP status codes, error text, and startup warnings for config validation failures
• •Fixed critical permission handling bug where approved actions were dropped when self-hosted runners restarted
• •Dynamic workflows now default to medium size guideline (fewer than 15 agents) with configurable sizing options
• •Enhanced stream-json nested subagent forwarding with `--forward-subagent-text` flag for depth-2+ subagents
• •Improved accessibility with screen-reader mode fix and Vim mode navigation improvements
• •Added structured failure categories for self-hosted runner spawn/session failures to distinguish hook errors, crashes, and config issues

Workflow Diagram

Concepts

Artifacts (2)

CLAUDE_CODE_MAX_SUBAGENT_SPAWN_DEPTH=1

sandbox:
  network:
    strictAllowlist: true