Unofficial AI-summarized news site (not affiliated with any AI company)
AI News JP / www.ai-news.jp
🔵 Standard AI Summary · Source: Claude Code Releases

11 bug fixes & security hardening. Skills now block `!` command execution locally—key for safe skill sync

Claude Code v2.1.228 Released: Session Stability and Security Enhancements

Original: v2.1.228

Importance: 運用安定性とセキュリティの改善が多数含まれるが、破壊的変更や新機能ではなくバグ修正・マイナーアップデート

Summary

Anthropic's Claude Code is updated to v2.1.228 with 11 major bug fixes and improvements. Key updates include fixing interactive sessions freezing after rare layout errors, Git not being found on Windows, /tui reverting the model incorrectly, and cross-session messaging failures after fresh install. Security enhancements include hardening skills synced from claude.ai (preventing shadowing local commands, sanitizing descriptions, disabling `!` command execution and `@` file expansion on local machines). Improved credential handling for Vertex AI, better Remote Control messaging, and refined Write/Edit tool behavior for newer models. Overall stability and security improvements for production users.

Key Points

  • Fixed session redraw freeze after rare layout error
  • Resolved Git detection failure on Windows
  • Patched Remote Control conversation history leak vulnerability
  • Hardened skill sync security (no local command execution)
  • Improved self-hosted runner reliability
View developer notes (APIs, breaking changes, migration)

v2.1.228 technical fixes: (1) Fixed interactive sessions completely stop redrawing after rare internal layout errors while process runs. (2) Windows Git detection failure when Claude Code launched from parent folder of git installation resolved. (3) /tui command now preserves /model changes across responses (no revert). (4) Cross-session messaging initialization fixed—no longer starts without inbox in first session post-install/upgrade. (5) Remote Control /resume credential leak patched—conversation title/history no longer exposed to connected session. (6) Self-hosted runners: checkout hook failures on un-pushed repositories now skip with warning instead of failing entire session. (7) Skills synced from claude.ai hardened: no longer shadow local commands or MCP prompts, descriptions sanitized & labeled, local execution blocks `!` command & `@` file expansion. (8) Write tool now matches Edit tool rules—newer models can overwrite unread files; older models still require read-first. (9) Vertex AI credentials: expired/missing Google Cloud credentials now fail within seconds vs. retry-for-minutes. (10) Session cleanup no longer deletes plugin memory folder contents. (11) Settings merge fixed: marketplace entries no longer inherit custom headers from lower-precedence tiers.

Claude Code障害Audience: 開発者Audience: 企業導入担当

Source: https://github.com/anthropics/claude-code/releases/tag/v2.1.228

Outlet: Claude Code Releases

This article is an AI-generated summary (OpenAI GPT-4o-mini) of publicly available information from Anthropic, OpenAI, Google, Meta, Mistral, DeepSeek, Sakana, and other vendors. The original source URL is always provided in accordance with fair-use citation requirements. Summaries are AI-generated and may contain mistranslations or misinterpretations. Always verify details with the original source.