AI HOT (Curated Pool)May 23, 2026, 06:09 UTC+8

v2.1.149 release summary

v2.1.149 更新摘要

Claude Code v2.1.149 adds categorized /usage reporting, an enterprise allowAllClaudeAiMcps setting for cloud MCP connectors, and fixes three security issues involving PowerShell permission bypass, Git worktree sandbox allowlist overflow, and otelHeadersHelper failures when script paths contain spaces.

Why it matters: Official Claude Code point release with concrete changes but limited blast radius: /usage categories, an enterprise MCP allow switch, and PowerShell bypass fixes hit developer security and governance needs.

Read the original ↗Export Markdown

Take

Claude Code v2.1.149 shows the old failure mode of tool agents: once the model touches local shell, Git worktrees, and telemetry scripts, execution boundaries break before model quality does. This release fixes PowerShell permission bypass, Git worktree sandbox allowlist overflow, and otelHeadersHelper failures on paths with spaces. All three sit in the run path, not the chat layer.

The categorized /usage view and enterprise allowAllClaudeAiMcps setting are procurement controls. The wild part is Anthropic pushing cloud MCP connectors while giving enterprises a broad allow switch. The faster MCP spreads, the more expensive default trust gets. Claude Code’s fight is no longer just better code generation; it is whether the agent’s system privileges can be caged tightly enough for enterprise machines.

Source: AI HOT (Curated Pool) · github.com
Published: May 23, 2026, 06:09 UTC+8 · May 23, 2026