30 June 2026claude-code#release#claude-code#security#subagents#sonnet
Claude Code in June 2026 — safe mode, nested agents, and Sonnet 5
June added safe mode, nested subagents, credential isolation, MCP login commands, managed version limits, and Sonnet 5 as the default model.
Claude Code 2.1.160 through 2.1.197 strengthened troubleshooting and policy while allowing agent hierarchies to grow deeper.
What changed
--safe-modedisabled instructions, plugins, skills, hooks, and MCP servers for clean troubleshooting.- Subagents gained the ability to spawn subagents, with depth controls added in later releases.
sandbox.credentialscould block sandboxed commands from reading credential files and secret environment variables.claude mcp loginandlogoutenabled MCP authentication from the CLI, including SSH-friendly flows.- Administrators could enforce minimum and maximum Claude Code versions.
- Shell startup files and sensitive Git configuration received stronger write prompts.
- Sonnet 5 became the default model with a native 1M-token context window and introductory API pricing.
Why it matters
Safe mode gives users a clean way to separate core-product failures from customization problems. Nested agents increase capability and cost together, so keep depth and concurrency bounded.