Claude Code v2.1.205: Transcript-Tampering Guard and Session Reliability Fixes
Anthropic
Released July 8, v2.1.205 introduces an auto-mode rule that blocks any attempt to tamper with session transcript files — a security hardening measure for unattended agents. Also fixes: `--json-schema` producing unstructured output on invalid schemas; messages silently dropped at the `--max-turns` limit; Windows worktree removal; session-to-PR linking improvements.
Why it matters
The transcript-tampering guard directly hardens autonomous agents against prompt injection via history manipulation. The max-turns message-loss fix is critical for long-running agentic tasks that previously silently dropped context near limits.
Importance: 2/5
Patch release with notable security hardening (transcript guard) for autonomous Claude Code sessions.