v2.1.172

Source: Anthropic Claude Code Date: 2026-06-10 URL: https://github.com/anthropics/claude-code/releases/tag/v2.1.172

Summary

Claude Code v2.1.172 ships two headline changes: sub-agents can now spawn their own sub-agents up to five levels deep, and a batch of multi-level correctness fixes for the agent panel, model picker, and permission-wildcard matching. On the infrastructure side: Amazon Bedrock gains proper AWS config-file region resolution, and a context-limit edge case (1M sessions without usage credits getting permanently stuck) is closed.

Implications

Feeds the agent-layer convergence thread directly. Recursive sub-agent spawning is a meaningful architectural shift — Claude Code is moving from a single-depth orchestrator to a tree-structured execution model, putting it on a design trajectory closer to Codex’s independent dispatch model even while keeping the steering-first UX. The wildcard permission fix (Read(secrets-*/config.json)) and the background-agent project-settings isolation fix both indicate the trust surface is expanding as depth increases, which is exactly where provenance and permission models will need to harden next.

• Watch: how deep nesting behaves in practice for long-running agentic tasks — five levels of concurrent sub-agents is a significant blast radius if a mid-tree node misfires.
• The Bedrock region-resolution fix signals continued investment in enterprise deployment paths alongside the claude.ai consumer product.