Claude Code release 2.1.301

Summary

Claude Code 2.1.301 is mostly a fixes release. It adds a cap on how long unattended retry mode waits for usage limits, OpenTelemetry fields that explain auto mode decisions, and a safer model-switch flow. Many fixes cover worktrees, search tools, and terminal rendering.

Key points

• CLAUDE_CODE_RETRY_WATCHDOG_MAX_USAGE_LIMIT_WAIT_MS: with unattended retry mode (CLAUDE_CODE_RETRY_WATCHDOG), ends a request instead of waiting when the usage limit resets too far in the future.
• OpenTelemetry tool_decision events gain classifier_verdict and permission_mode, which separate auto mode classifier decisions from permission rules, and classifier_category (with OTEL_LOG_TOOL_DETAILS=1) for the rule a block matched.
• Switching to a model with a smaller context window now summarises the conversation with the previous model first, and no longer spends usage credits on that without asking.
• The fable and best aliases fall back once to the previous Fable model when the newest is not available to an organisation.
• Quitting no longer silently deletes a hook-created worktree that may hold unpushed commits.
• Grep and Glob now report an error when the search program fails, instead of "no matches"; the Read tool no longer reads binary formats such as .parquet as text.
• A fix stops repeated compaction when an LLM gateway reports a different model name than the one requested.
• VS Code extension renders LaTeX maths; vim mode gains the % motion.

Why it matters

Silent "no matches" results and lost worktree commits are the kind of quiet failures that mislead an agent or lose work. The new telemetry fields make it possible to audit why auto mode allowed or blocked a tool call.