Claude Code v2.1.301 adds retry watchdog cap, OTEL classifier fields, vim % motion
cliClaude Code 2.1.301 ships a new `CLAUDE_CODE_RETRY_WATCHDOG_MAX_USAGE_LIMIT_WAIT_MS` env var so unattended retry mode aborts when a usage-limit reset is too far away, adds `classifier_verdict`, `permission_mode`, and `classifier_category` fields to OpenTelemetry `tool_decision` events, documents OTEL tracing in the `claude-api` skill, plus many fixes: smarter cross-model context handoff, binary-file Read guards, vim-mode polish, and VS Code LaTeX rendering.
An agent can now cap how long unattended Claude Code waits at a usage limit before failing, and distinguish auto-mode classifier verdicts from permission-rule decisions in OpenTelemetry traces.
Point release; headline capabilities are the retry-watchdog timeout env var and expanded OTEL tool_decision telemetry (classifier_verdict/category, permission_mode) useful for agent observability.
npm i -g @anthropic-ai/claude-code@2.1.301npm i -g @anthropic-ai/claude-code@2.1.301
# Bound how long unattended retry mode waits at a usage limit
export CLAUDE_CODE_RETRY_WATCHDOG=1
export CLAUDE_CODE_RETRY_WATCHDOG_MAX_USAGE_LIMIT_WAIT_MS=600000 # 10 min
# Richer OTEL tool_decision events
export OTEL_LOG_TOOL_DETAILS=1
claude -p "refactor the auth module and run tests"