Security hardening + cache stability are the core. Feels like focused operational bug-squashing.
Claude Code v2.1.248: Security Hardening and Stability Improvements
Original: v2.1.248
Importance: 多くのユーザーに影響する安定性向上とセキュリティ強化だが、破壊的変更なく段階的機能追加のため。
Summary
Anthropic released Claude Code v2.1.248 with a restricted mode (`--restricted`) that disables command execution and web-fetch tools for enhanced security. Added `/usage-credits` for Enterprise users to request higher usage limits via AWS Marketplace or Vertex. Fixed prompt-cache misses in long sessions caused by tool re-rendering after OAuth refresh, session resurrection bugs on Windows, and GitHub CLI token validation warnings. Improved cross-session messaging on Bedrock/Vertex/Foundry and addressed transcript cleanup for long-retained desktop sessions.
Key Points
- `--restricted` flag enables permission control
- Fixed prompt-cache loss in long sessions
- Improved stability on OAuth token refresh
- Multiple Windows UX and keyboard input fixes
- Enterprise usage-limit request feature added
View developer notes (APIs, breaking changes, migration)
v2.1.248 adds `--restricted` flag (env `CLAUDE_CODE_RESTRICTED=1`) disabling command execution, WebFetch, out-of-dir file access, and `bypassPermissions`. Agent frontmatter's `experimental.cacheTtl` ('5m'/'1h') controls prompt cache TTL when no subagent setting exists. `claude self-hosted-runner --client-label <label>` overrides runner registration label. Fixes tool-definition re-render causing prompt-cache loss post-OAuth refresh, `ScheduleWakeup` mismatch on usage overage, 30-day desktop session cleanup, Windows CLI keyboard input, malformed PR-status cache. Implements cross-session messaging (SendMessage/ListAgents) on Bedrock, Vertex, Foundry when telemetry disabled.
Source: https://github.com/anthropics/claude-code/releases/tag/v2.1.248
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.