v2.1.178

Source: Anthropic Claude Code Date: 2026-06-15 URL: https://github.com/anthropics/claude-code/releases/tag/v2.1.178

Summary

Claude Code v2.1.178 (released 2026-06-15) adds Tool(param:value) syntax for permission rules, enabling parameter-level matching (e.g. blocking only Opus subagents). Nested .claude/ directory resolution now uses proximity rules — closest config wins on name collision. Subagent evaluation tightens: the auto-mode classifier now reviews subagent spawns before launch, closing a gap where a blocked action could be requested without review.

Implications

• Agentic engineering patterns: parameter-scoped permission rules materially change how multi-agent systems are governed — operators can now whitelist tool use by model tier or input value, not just by tool name. This is a meaningful capability for any pipeline that runs mixed-model subagent fleets.
• Developer tooling: nested .claude/ directory resolution enables per-workspace skill and config layering; useful for monorepos or multi-project setups.
• OpenAI/Anthropic distribution strategy: classifier-gated subagent spawns and improved Remote Control error surfacing suggest Anthropic is investing in the agentic runtime as a differentiator, not just the underlying model.