2 min read AI-generated

Claude Code 2.1.119/120: Eight Regressions in 24 Hours

Copy article as Markdown

Two patch releases, eight bugs: Claude Code had a rough weekend. Resume crashes, CLAUDE.md gets ignored, and the community recommends rolling back to 2.1.117.

Featured image for "Claude Code 2.1.119/120: Eight Regressions in 24 Hours"

If you’re using Claude Code and wondered why everything broke over the weekend — you’re not alone. Anthropic shipped two patch releases within 24 hours on April 23-24, and both introduced more problems than they solved.

What Happened

Version 2.1.119 brought genuinely useful features: persistent config settings, GitLab and Bitbucket support for --from-pr, and PowerShell auto-approval. Sounds great. The problem: version 2.1.120 followed a day later, and together the two releases introduced eight documented regressions.

The Bug List

The community compiled a ‘Survival Checklist’ on GitHub, and it’s rough:

  • Resume/Continue crashes: claude --resume and claude --continue crash on startup with a JavaScript error (‘g9H is not a function’). Anyone trying to continue their sessions was stuck.
  • CLAUDE.md ignored: Project conventions, tool inventories, design policies — all invisible. One of the most-used features, silently broken.
  • Sandbox errors: On macOS, 2.1.120 refuses to start a resumed session with ‘sandbox required but unavailable’.
  • MCP connectors lost: Web sessions lost their claude.ai MCP connectors.
  • UI duplication: The same text appears at two positions simultaneously.
  • WSL2 freeze: /mcp completely freezes under Windows Subsystem for Linux.
  • Auto-update break: The automatic update mechanism itself was broken.
  • VS Code timing issue: The VS Code extension was published before the CLI was available.

Community Response

The community’s recommendation is clear: roll back to version 2.1.117. That was the last stable version after the quality fix from April 20. Anthropic at least adjusted the auto-updater so affected clients on 2.1.120 are automatically downgraded to 2.1.119.

Context: The Quality Saga Continues

The timing is particularly unfortunate. Just on April 23, Anthropic had published a detailed postmortem acknowledging three separate quality issues in Claude Code — including a reasoning effort bug and a cache problem that had been causing frustration since early March. The message was: ‘We understand, it’s fixed.’ And then two days later, eight new bugs.

I want to be fair here: fast release cycles mean things sometimes break. But the pattern over the past few weeks shows that quality assurance at Claude Code still has room to improve. For a tool that developers trust with their code, stability isn’t a nice-to-have.

Sources: