2026-05-15 · Anthropic

v2.1.143

agentsmodelsinfrastructure

read at source ↗ github.com

v2.1.143

Source: Anthropic Claude Code Date: 2026-05-15 URL: https://github.com/anthropics/claude-code/releases/tag/v2.1.143

Summary

Claude Code v2.1.143 is a focused platform-health release: plugin dependency enforcement arrives (disabling a plugin that other enabled plugins depend on is now blocked, with a chain-hint), background sessions gain model and effort persistence across idle wakes, and claude agents picks up the full flag set (--add-dir, --model, --effort, --permission-mode, etc.). PowerShell on Windows now passes -ExecutionPolicy Bypass by default, unblocking Codex CLI for Bedrock/Vertex/Foundry users who had been hitting policy restrictions. A cluster of crash and hang fixes covers corrupted credentials startup hangs, Full Disk Access failures for background macOS jobs, and stop hooks that were looping past their cap.

Implications

  • Agent infrastructure maturation. The dependency graph for plugins, persistent model/effort state across background sessions, and direct worktree editing mode (bgIsolation: "none") are all plumbing that makes background agent workflows more composable. This is incremental hardening, not a capability leap, but it reduces friction for teams running multi-agent setups that mix foreground and background sessions.
  • Windows parity. The PowerShell execution policy default and claude agents flag parity close gaps that were disproportionately felt by enterprise Windows users. Codex CLI’s viability on Windows is a prerequisite for broad enterprise deployment; this release moves that needle.
  • Reliability as a competitive signal. Fixing startup hangs from corrupted credentials and macOS Full Disk Access failures are table-stakes reliability items. At the velocity Claude Code ships, patch cadence itself is part of the product surface — enterprises evaluating agentic coding tools treat “does it crash on Monday morning?” as a signal.

← all signals