Claude Code v2.1.219 updates default model to Opus 5, adds nested subagent forwarding and workflow size controls

Anthropic released version 2.1.219 of claude-code, its command-line coding agent, introducing several developer-facing updates. The default model is now Claude Opus 5, which features a 1M context window and offers a fast mode priced at $10 per million tokens input and $50 per million tokens output.

Nested subagent forwarding has been increased to a default depth of 3, enabling deeper multi-agent workflows without additional configuration. Teams can restrict nesting by setting CLAUDE_CODE_MAX_SUBAGENT_SPAWN_DEPTH to 1.

Dynamic workflow size guidelines now default to a medium setting aimed at fewer than 15 agents, with the current size displayed in the running-workflow status line and adjustable via the /config interface.

For MCP integrations, the release improves error visibility by surfacing HTTP status codes and error text when servers fail to connect, and warns when MCP config values contain hidden leading or trailing whitespace. The claude mcp list and /mcp commands now expose these details directly.

Self-hosted runner diagnostics have been refined with structured failure categories for hook errors, runner crashes, and config errors, improving clarity during troubleshooting. Runner lifecycle issues, such as stale active rows after SIGTERM, have also been addressed to ensure clean deregistration.

Additional fixes include correct terminal copy behavior in GNU screen, improved Vim mode navigation, and clearer error messages for Remote Control settings. The claude-api skill has been updated to default to Opus 5, with a migration path from Opus 4.8.